You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
rocksdb/db/db_test2.cc

7576 lines
269 KiB

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#include <atomic>
#include <cstdlib>
#include <functional>
#include <memory>
#include "db/db_test_util.h"
#include "db/read_callback.h"
#include "options/options_helper.h"
#include "port/port.h"
#include "port/stack_trace.h"
Add manifest fix-up utility for file temperatures (#9683) Summary: The goal of this change is to allow changes to the "current" (in FileSystem) file temperatures to feed back into DB metadata, so that they can inform decisions and stats reporting. In part because of modular code factoring, it doesn't seem easy to do this automagically, where opening an SST file and observing current Temperature different from expected would trigger a change in metadata and DB manifest write (essentially giving the deep read path access to the write path). It is also difficult to do this while the DB is open because of the limitations of LogAndApply. This change allows updating file temperature metadata on a closed DB using an experimental utility function UpdateManifestForFilesState() or `ldb update_manifest --update_temperatures`. This should suffice for "migration" scenarios where outside tooling has placed or re-arranged DB files into a (different) tiered configuration without going through RocksDB itself (currently, only compaction can change temperature metadata). Some details: * Refactored and added unit test for `ldb unsafe_remove_sst_file` because of shared functionality * Pulled in autovector.h changes from https://github.com/facebook/rocksdb/issues/9546 to fix SuperVersionContext move constructor (related to an older draft of this change) Possible follow-up work: * Support updating manifest with file checksums, such as when a new checksum function is used and want existing DB metadata updated for it. * It's possible that for some repair scenarios, lighter weight than full repair, we might want to support UpdateManifestForFilesState() to modify critical file details like size or checksum using same algorithm. But let's make sure these are differentiated from modifying file details in ways that don't suspect corruption (or require extreme trust). Pull Request resolved: https://github.com/facebook/rocksdb/pull/9683 Test Plan: unit tests added Reviewed By: jay-zhuang Differential Revision: D34798828 Pulled By: pdillinger fbshipit-source-id: cfd83e8fb10761d8c9e7f9c020d68c9106a95554
3 years ago
#include "rocksdb/experimental.h"
#include "rocksdb/iostats_context.h"
#include "rocksdb/persistent_cache.h"
#include "rocksdb/trace_record.h"
#include "rocksdb/trace_record_result.h"
#include "rocksdb/utilities/replayer.h"
#include "rocksdb/wal_filter.h"
#include "test_util/testutil.h"
#include "util/random.h"
#include "utilities/fault_injection_env.h"
namespace ROCKSDB_NAMESPACE {
class DBTest2 : public DBTestBase {
public:
DBTest2() : DBTestBase("db_test2", /*env_do_fsync=*/true) {}
};
#ifndef ROCKSDB_LITE
TEST_F(DBTest2, OpenForReadOnly) {
DB* db_ptr = nullptr;
std::string dbname = test::PerThreadDBPath("db_readonly");
Options options = CurrentOptions();
options.create_if_missing = true;
// OpenForReadOnly should fail but will create <dbname> in the file system
ASSERT_NOK(DB::OpenForReadOnly(options, dbname, &db_ptr));
// Since <dbname> is created, we should be able to delete the dir
// We first get the list files under <dbname>
// There should not be any subdirectories -- this is not checked here
std::vector<std::string> files;
ASSERT_OK(env_->GetChildren(dbname, &files));
for (auto& f : files) {
ASSERT_OK(env_->DeleteFile(dbname + "/" + f));
}
// <dbname> should be empty now and we should be able to delete it
ASSERT_OK(env_->DeleteDir(dbname));
options.create_if_missing = false;
// OpenForReadOnly should fail since <dbname> was successfully deleted
ASSERT_NOK(DB::OpenForReadOnly(options, dbname, &db_ptr));
// With create_if_missing false, there should not be a dir in the file system
ASSERT_NOK(env_->FileExists(dbname));
}
TEST_F(DBTest2, OpenForReadOnlyWithColumnFamilies) {
DB* db_ptr = nullptr;
std::string dbname = test::PerThreadDBPath("db_readonly");
Options options = CurrentOptions();
options.create_if_missing = true;
ColumnFamilyOptions cf_options(options);
std::vector<ColumnFamilyDescriptor> column_families;
column_families.push_back(
ColumnFamilyDescriptor(kDefaultColumnFamilyName, cf_options));
column_families.push_back(ColumnFamilyDescriptor("goku", cf_options));
std::vector<ColumnFamilyHandle*> handles;
// OpenForReadOnly should fail but will create <dbname> in the file system
ASSERT_NOK(
DB::OpenForReadOnly(options, dbname, column_families, &handles, &db_ptr));
// Since <dbname> is created, we should be able to delete the dir
// We first get the list files under <dbname>
// There should not be any subdirectories -- this is not checked here
std::vector<std::string> files;
ASSERT_OK(env_->GetChildren(dbname, &files));
for (auto& f : files) {
ASSERT_OK(env_->DeleteFile(dbname + "/" + f));
}
// <dbname> should be empty now and we should be able to delete it
ASSERT_OK(env_->DeleteDir(dbname));
options.create_if_missing = false;
// OpenForReadOnly should fail since <dbname> was successfully deleted
ASSERT_NOK(
DB::OpenForReadOnly(options, dbname, column_families, &handles, &db_ptr));
// With create_if_missing false, there should not be a dir in the file system
ASSERT_NOK(env_->FileExists(dbname));
}
class TestReadOnlyWithCompressedCache
: public DBTestBase,
public testing::WithParamInterface<std::tuple<int, bool>> {
public:
TestReadOnlyWithCompressedCache()
: DBTestBase("test_readonly_with_compressed_cache",
/*env_do_fsync=*/true) {
max_open_files_ = std::get<0>(GetParam());
use_mmap_ = std::get<1>(GetParam());
}
int max_open_files_;
bool use_mmap_;
};
TEST_P(TestReadOnlyWithCompressedCache, ReadOnlyWithCompressedCache) {
if (use_mmap_ && !IsMemoryMappedAccessSupported()) {
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
ROCKSDB_GTEST_SKIP("Test requires MMAP support");
return;
}
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Put("foo2", "barbarbarbarbarbarbarbar"));
ASSERT_OK(Flush());
DB* db_ptr = nullptr;
Options options = CurrentOptions();
options.allow_mmap_reads = use_mmap_;
options.max_open_files = max_open_files_;
options.compression = kSnappyCompression;
BlockBasedTableOptions table_options;
table_options.block_cache_compressed = NewLRUCache(8 * 1024 * 1024);
table_options.no_block_cache = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.statistics = CreateDBStatistics();
ASSERT_OK(DB::OpenForReadOnly(options, dbname_, &db_ptr));
std::string v;
ASSERT_OK(db_ptr->Get(ReadOptions(), "foo", &v));
ASSERT_EQ("bar", v);
ASSERT_EQ(0, options.statistics->getTickerCount(BLOCK_CACHE_COMPRESSED_HIT));
ASSERT_OK(db_ptr->Get(ReadOptions(), "foo", &v));
ASSERT_EQ("bar", v);
if (Snappy_Supported()) {
if (use_mmap_) {
ASSERT_EQ(0,
options.statistics->getTickerCount(BLOCK_CACHE_COMPRESSED_HIT));
} else {
ASSERT_EQ(1,
options.statistics->getTickerCount(BLOCK_CACHE_COMPRESSED_HIT));
}
}
delete db_ptr;
}
INSTANTIATE_TEST_CASE_P(TestReadOnlyWithCompressedCache,
TestReadOnlyWithCompressedCache,
::testing::Combine(::testing::Values(-1, 100),
::testing::Bool()));
class PartitionedIndexTestListener : public EventListener {
public:
void OnFlushCompleted(DB* /*db*/, const FlushJobInfo& info) override {
ASSERT_GT(info.table_properties.index_partitions, 1);
ASSERT_EQ(info.table_properties.index_key_is_user_key, 0);
}
};
TEST_F(DBTest2, PartitionedIndexUserToInternalKey) {
const int kValueSize = 10500;
const int kNumEntriesPerFile = 1000;
const int kNumFiles = 3;
const int kNumDistinctKeys = 30;
BlockBasedTableOptions table_options;
Options options = CurrentOptions();
options.disable_auto_compactions = true;
table_options.index_type = BlockBasedTableOptions::kTwoLevelIndexSearch;
PartitionedIndexTestListener* listener = new PartitionedIndexTestListener();
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.listeners.emplace_back(listener);
std::vector<const Snapshot*> snapshots;
Reopen(options);
Random rnd(301);
for (int i = 0; i < kNumFiles; i++) {
for (int j = 0; j < kNumEntriesPerFile; j++) {
int key_id = (i * kNumEntriesPerFile + j) % kNumDistinctKeys;
std::string value = rnd.RandomString(kValueSize);
ASSERT_OK(Put("keykey_" + std::to_string(key_id), value));
snapshots.push_back(db_->GetSnapshot());
}
ASSERT_OK(Flush());
}
for (auto s : snapshots) {
db_->ReleaseSnapshot(s);
}
}
#endif // ROCKSDB_LITE
class PrefixFullBloomWithReverseComparator
: public DBTestBase,
public ::testing::WithParamInterface<bool> {
public:
PrefixFullBloomWithReverseComparator()
: DBTestBase("prefix_bloom_reverse", /*env_do_fsync=*/true) {}
void SetUp() override { if_cache_filter_ = GetParam(); }
bool if_cache_filter_;
};
TEST_P(PrefixFullBloomWithReverseComparator,
PrefixFullBloomWithReverseComparator) {
Options options = last_options_;
options.comparator = ReverseBytewiseComparator();
options.prefix_extractor.reset(NewCappedPrefixTransform(3));
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
BlockBasedTableOptions bbto;
if (if_cache_filter_) {
bbto.no_block_cache = false;
bbto.cache_index_and_filter_blocks = true;
bbto.block_cache = NewLRUCache(1);
}
bbto.filter_policy.reset(NewBloomFilterPolicy(10, false));
bbto.whole_key_filtering = false;
options.table_factory.reset(NewBlockBasedTableFactory(bbto));
DestroyAndReopen(options);
ASSERT_OK(dbfull()->Put(WriteOptions(), "bar123", "foo"));
ASSERT_OK(dbfull()->Put(WriteOptions(), "bar234", "foo2"));
ASSERT_OK(dbfull()->Put(WriteOptions(), "foo123", "foo3"));
ASSERT_OK(dbfull()->Flush(FlushOptions()));
if (bbto.block_cache) {
bbto.block_cache->EraseUnRefEntries();
}
std::unique_ptr<Iterator> iter(db_->NewIterator(ReadOptions()));
iter->Seek("bar345");
ASSERT_OK(iter->status());
ASSERT_TRUE(iter->Valid());
ASSERT_EQ("bar234", iter->key().ToString());
ASSERT_EQ("foo2", iter->value().ToString());
iter->Next();
ASSERT_TRUE(iter->Valid());
ASSERT_EQ("bar123", iter->key().ToString());
ASSERT_EQ("foo", iter->value().ToString());
iter->Seek("foo234");
ASSERT_OK(iter->status());
ASSERT_TRUE(iter->Valid());
ASSERT_EQ("foo123", iter->key().ToString());
ASSERT_EQ("foo3", iter->value().ToString());
iter->Seek("bar");
ASSERT_OK(iter->status());
ASSERT_TRUE(!iter->Valid());
}
INSTANTIATE_TEST_CASE_P(PrefixFullBloomWithReverseComparator,
PrefixFullBloomWithReverseComparator, testing::Bool());
TEST_F(DBTest2, IteratorPropertyVersionNumber) {
ASSERT_OK(Put("", ""));
Iterator* iter1 = db_->NewIterator(ReadOptions());
ASSERT_OK(iter1->status());
std::string prop_value;
ASSERT_OK(
iter1->GetProperty("rocksdb.iterator.super-version-number", &prop_value));
uint64_t version_number1 =
static_cast<uint64_t>(std::atoi(prop_value.c_str()));
ASSERT_OK(Put("", ""));
ASSERT_OK(Flush());
Iterator* iter2 = db_->NewIterator(ReadOptions());
ASSERT_OK(iter2->status());
ASSERT_OK(
iter2->GetProperty("rocksdb.iterator.super-version-number", &prop_value));
uint64_t version_number2 =
static_cast<uint64_t>(std::atoi(prop_value.c_str()));
ASSERT_GT(version_number2, version_number1);
ASSERT_OK(Put("", ""));
Iterator* iter3 = db_->NewIterator(ReadOptions());
ASSERT_OK(iter3->status());
ASSERT_OK(
iter3->GetProperty("rocksdb.iterator.super-version-number", &prop_value));
uint64_t version_number3 =
static_cast<uint64_t>(std::atoi(prop_value.c_str()));
ASSERT_EQ(version_number2, version_number3);
iter1->SeekToFirst();
ASSERT_OK(
iter1->GetProperty("rocksdb.iterator.super-version-number", &prop_value));
uint64_t version_number1_new =
static_cast<uint64_t>(std::atoi(prop_value.c_str()));
ASSERT_EQ(version_number1, version_number1_new);
delete iter1;
delete iter2;
delete iter3;
}
TEST_F(DBTest2, CacheIndexAndFilterWithDBRestart) {
Options options = CurrentOptions();
options.create_if_missing = true;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
BlockBasedTableOptions table_options;
table_options.cache_index_and_filter_blocks = true;
table_options.filter_policy.reset(NewBloomFilterPolicy(20));
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
CreateAndReopenWithCF({"pikachu"}, options);
ASSERT_OK(Put(1, "a", "begin"));
ASSERT_OK(Put(1, "z", "end"));
ASSERT_OK(Flush(1));
TryReopenWithColumnFamilies({"default", "pikachu"}, options);
std::string value;
value = Get(1, "a");
}
TEST_F(DBTest2, MaxSuccessiveMergesChangeWithDBRecovery) {
Options options = CurrentOptions();
options.create_if_missing = true;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
options.max_successive_merges = 3;
options.merge_operator = MergeOperators::CreatePutOperator();
options.disable_auto_compactions = true;
DestroyAndReopen(options);
ASSERT_OK(Put("poi", "Finch"));
ASSERT_OK(db_->Merge(WriteOptions(), "poi", "Reese"));
ASSERT_OK(db_->Merge(WriteOptions(), "poi", "Shaw"));
ASSERT_OK(db_->Merge(WriteOptions(), "poi", "Root"));
options.max_successive_merges = 2;
Reopen(options);
}
#ifndef ROCKSDB_LITE
class DBTestSharedWriteBufferAcrossCFs
: public DBTestBase,
public testing::WithParamInterface<std::tuple<bool, bool>> {
public:
DBTestSharedWriteBufferAcrossCFs()
: DBTestBase("db_test_shared_write_buffer", /*env_do_fsync=*/true) {}
void SetUp() override {
use_old_interface_ = std::get<0>(GetParam());
cost_cache_ = std::get<1>(GetParam());
}
bool use_old_interface_;
bool cost_cache_;
};
TEST_P(DBTestSharedWriteBufferAcrossCFs, SharedWriteBufferAcrossCFs) {
Options options = CurrentOptions();
options.arena_block_size = 4096;
auto flush_listener = std::make_shared<FlushCounterListener>();
options.listeners.push_back(flush_listener);
// Don't trip the listener at shutdown.
options.avoid_flush_during_shutdown = true;
// Avoid undeterministic value by malloc_usable_size();
// Force arena block size to 1
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"Arena::Arena:0", [&](void* arg) {
size_t* block_size = static_cast<size_t*>(arg);
*block_size = 1;
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"Arena::AllocateNewBlock:0", [&](void* arg) {
std::pair<size_t*, size_t*>* pair =
static_cast<std::pair<size_t*, size_t*>*>(arg);
*std::get<0>(*pair) = *std::get<1>(*pair);
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
// The total soft write buffer size is about 105000
std::shared_ptr<Cache> cache = NewLRUCache(4 * 1024 * 1024, 2);
ASSERT_LT(cache->GetUsage(), 256 * 1024);
if (use_old_interface_) {
options.db_write_buffer_size = 120000; // this is the real limit
} else if (!cost_cache_) {
options.write_buffer_manager.reset(new WriteBufferManager(114285));
} else {
options.write_buffer_manager.reset(new WriteBufferManager(114285, cache));
}
options.write_buffer_size = 500000; // this is never hit
CreateAndReopenWithCF({"pikachu", "dobrynia", "nikitich"}, options);
WriteOptions wo;
wo.disableWAL = true;
std::function<void()> wait_flush = [&]() {
ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable(handles_[0]));
ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable(handles_[1]));
ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable(handles_[2]));
ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable(handles_[3]));
Fix test race conditions with OnFlushCompleted() (#9617) Summary: We often see flaky tests due to `DB::Flush()` or `DBImpl::TEST_WaitForFlushMemTable()` not waiting until event listeners complete. For example, https://github.com/facebook/rocksdb/issues/9084, https://github.com/facebook/rocksdb/issues/9400, https://github.com/facebook/rocksdb/issues/9528, plus two new ones this week: "EventListenerTest.OnSingleDBFlushTest" and "DBFlushTest.FireOnFlushCompletedAfterCommittedResult". I ran a `make check` with the below race condition-coercing patch and fixed issues it found besides old BlobDB. ``` diff --git a/db/db_impl/db_impl_compaction_flush.cc b/db/db_impl/db_impl_compaction_flush.cc index 0e1864788..aaba68c4a 100644 --- a/db/db_impl/db_impl_compaction_flush.cc +++ b/db/db_impl/db_impl_compaction_flush.cc @@ -861,6 +861,8 @@ void DBImpl::NotifyOnFlushCompleted( mutable_cf_options.level0_stop_writes_trigger); // release lock while notifying events mutex_.Unlock(); + bg_cv_.SignalAll(); + sleep(1); { for (auto& info : *flush_jobs_info) { info->triggered_writes_slowdown = triggered_writes_slowdown; ``` The reason I did not fix old BlobDB issues is because it appears to have a fundamental (non-test) issue. In particular, it uses an EventListener to keep track of the files. OnFlushCompleted() could be delayed until even after a compaction involving that flushed file completes, causing the compaction to unexpectedly delete an untracked file. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9617 Test Plan: `make check` including the race condition coercing patch Reviewed By: hx235 Differential Revision: D34384022 Pulled By: ajkr fbshipit-source-id: 2652ded39b415277c5d6a628414345223930514e
3 years ago
// Ensure background work is fully finished including listener callbacks
// before accessing listener state.
ASSERT_OK(dbfull()->TEST_WaitForBackgroundWork());
};
// Create some data and flush "default" and "nikitich" so that they
// are newer CFs created.
flush_listener->expected_flush_reason = FlushReason::kManualFlush;
ASSERT_OK(Put(3, Key(1), DummyString(1), wo));
Flush(3);
ASSERT_OK(Put(3, Key(1), DummyString(1), wo));
ASSERT_OK(Put(0, Key(1), DummyString(1), wo));
Flush(0);
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
static_cast<uint64_t>(1));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "nikitich"),
static_cast<uint64_t>(1));
flush_listener->expected_flush_reason = FlushReason::kWriteBufferManager;
ASSERT_OK(Put(3, Key(1), DummyString(30000), wo));
if (cost_cache_) {
ASSERT_GE(cache->GetUsage(), 256 * 1024);
ASSERT_LE(cache->GetUsage(), 2 * 256 * 1024);
}
wait_flush();
ASSERT_OK(Put(0, Key(1), DummyString(60000), wo));
if (cost_cache_) {
ASSERT_GE(cache->GetUsage(), 256 * 1024);
ASSERT_LE(cache->GetUsage(), 2 * 256 * 1024);
}
wait_flush();
ASSERT_OK(Put(2, Key(1), DummyString(1), wo));
// No flush should trigger
wait_flush();
{
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
static_cast<uint64_t>(1));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "pikachu"),
static_cast<uint64_t>(0));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "dobrynia"),
static_cast<uint64_t>(0));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "nikitich"),
static_cast<uint64_t>(1));
}
// Trigger a flush. Flushing "nikitich".
ASSERT_OK(Put(3, Key(2), DummyString(30000), wo));
wait_flush();
ASSERT_OK(Put(0, Key(1), DummyString(1), wo));
wait_flush();
{
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
static_cast<uint64_t>(1));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "pikachu"),
static_cast<uint64_t>(0));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "dobrynia"),
static_cast<uint64_t>(0));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "nikitich"),
static_cast<uint64_t>(2));
}
// Without hitting the threshold, no flush should trigger.
ASSERT_OK(Put(2, Key(1), DummyString(30000), wo));
wait_flush();
ASSERT_OK(Put(2, Key(1), DummyString(1), wo));
wait_flush();
ASSERT_OK(Put(2, Key(1), DummyString(1), wo));
wait_flush();
{
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
static_cast<uint64_t>(1));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "pikachu"),
static_cast<uint64_t>(0));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "dobrynia"),
static_cast<uint64_t>(0));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "nikitich"),
static_cast<uint64_t>(2));
}
// Hit the write buffer limit again. "default"
// will have been flushed.
ASSERT_OK(Put(2, Key(2), DummyString(10000), wo));
wait_flush();
ASSERT_OK(Put(3, Key(1), DummyString(1), wo));
wait_flush();
ASSERT_OK(Put(0, Key(1), DummyString(1), wo));
wait_flush();
ASSERT_OK(Put(0, Key(1), DummyString(1), wo));
wait_flush();
ASSERT_OK(Put(0, Key(1), DummyString(1), wo));
wait_flush();
{
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
static_cast<uint64_t>(2));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "pikachu"),
static_cast<uint64_t>(0));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "dobrynia"),
static_cast<uint64_t>(0));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "nikitich"),
static_cast<uint64_t>(2));
}
// Trigger another flush. This time "dobrynia". "pikachu" should not
// be flushed, althrough it was never flushed.
ASSERT_OK(Put(1, Key(1), DummyString(1), wo));
wait_flush();
ASSERT_OK(Put(2, Key(1), DummyString(80000), wo));
wait_flush();
ASSERT_OK(Put(1, Key(1), DummyString(1), wo));
wait_flush();
ASSERT_OK(Put(2, Key(1), DummyString(1), wo));
wait_flush();
{
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
static_cast<uint64_t>(2));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "pikachu"),
static_cast<uint64_t>(0));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "dobrynia"),
static_cast<uint64_t>(1));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "nikitich"),
static_cast<uint64_t>(2));
}
if (cost_cache_) {
ASSERT_GE(cache->GetUsage(), 256 * 1024);
Close();
options.write_buffer_manager.reset();
last_options_.write_buffer_manager.reset();
ASSERT_LT(cache->GetUsage(), 256 * 1024);
}
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
INSTANTIATE_TEST_CASE_P(DBTestSharedWriteBufferAcrossCFs,
DBTestSharedWriteBufferAcrossCFs,
::testing::Values(std::make_tuple(true, false),
std::make_tuple(false, false),
std::make_tuple(false, true)));
TEST_F(DBTest2, SharedWriteBufferLimitAcrossDB) {
std::string dbname2 = test::PerThreadDBPath("db_shared_wb_db2");
Options options = CurrentOptions();
options.arena_block_size = 4096;
auto flush_listener = std::make_shared<FlushCounterListener>();
options.listeners.push_back(flush_listener);
// Don't trip the listener at shutdown.
options.avoid_flush_during_shutdown = true;
// Avoid undeterministic value by malloc_usable_size();
// Force arena block size to 1
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"Arena::Arena:0", [&](void* arg) {
size_t* block_size = static_cast<size_t*>(arg);
*block_size = 1;
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"Arena::AllocateNewBlock:0", [&](void* arg) {
std::pair<size_t*, size_t*>* pair =
static_cast<std::pair<size_t*, size_t*>*>(arg);
*std::get<0>(*pair) = *std::get<1>(*pair);
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
options.write_buffer_size = 500000; // this is never hit
// Use a write buffer total size so that the soft limit is about
// 105000.
options.write_buffer_manager.reset(new WriteBufferManager(120000));
CreateAndReopenWithCF({"cf1", "cf2"}, options);
ASSERT_OK(DestroyDB(dbname2, options));
DB* db2 = nullptr;
ASSERT_OK(DB::Open(options, dbname2, &db2));
WriteOptions wo;
wo.disableWAL = true;
std::function<void()> wait_flush = [&]() {
ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable(handles_[0]));
ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable(handles_[1]));
ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable(handles_[2]));
ASSERT_OK(static_cast<DBImpl*>(db2)->TEST_WaitForFlushMemTable());
Fix test race conditions with OnFlushCompleted() (#9617) Summary: We often see flaky tests due to `DB::Flush()` or `DBImpl::TEST_WaitForFlushMemTable()` not waiting until event listeners complete. For example, https://github.com/facebook/rocksdb/issues/9084, https://github.com/facebook/rocksdb/issues/9400, https://github.com/facebook/rocksdb/issues/9528, plus two new ones this week: "EventListenerTest.OnSingleDBFlushTest" and "DBFlushTest.FireOnFlushCompletedAfterCommittedResult". I ran a `make check` with the below race condition-coercing patch and fixed issues it found besides old BlobDB. ``` diff --git a/db/db_impl/db_impl_compaction_flush.cc b/db/db_impl/db_impl_compaction_flush.cc index 0e1864788..aaba68c4a 100644 --- a/db/db_impl/db_impl_compaction_flush.cc +++ b/db/db_impl/db_impl_compaction_flush.cc @@ -861,6 +861,8 @@ void DBImpl::NotifyOnFlushCompleted( mutable_cf_options.level0_stop_writes_trigger); // release lock while notifying events mutex_.Unlock(); + bg_cv_.SignalAll(); + sleep(1); { for (auto& info : *flush_jobs_info) { info->triggered_writes_slowdown = triggered_writes_slowdown; ``` The reason I did not fix old BlobDB issues is because it appears to have a fundamental (non-test) issue. In particular, it uses an EventListener to keep track of the files. OnFlushCompleted() could be delayed until even after a compaction involving that flushed file completes, causing the compaction to unexpectedly delete an untracked file. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9617 Test Plan: `make check` including the race condition coercing patch Reviewed By: hx235 Differential Revision: D34384022 Pulled By: ajkr fbshipit-source-id: 2652ded39b415277c5d6a628414345223930514e
3 years ago
// Ensure background work is fully finished including listener callbacks
// before accessing listener state.
ASSERT_OK(dbfull()->TEST_WaitForBackgroundWork());
ASSERT_OK(
static_cast_with_check<DBImpl>(db2)->TEST_WaitForBackgroundWork());
};
// Trigger a flush on cf2
flush_listener->expected_flush_reason = FlushReason::kWriteBufferManager;
ASSERT_OK(Put(2, Key(1), DummyString(70000), wo));
wait_flush();
ASSERT_OK(Put(0, Key(1), DummyString(20000), wo));
wait_flush();
// Insert to DB2
ASSERT_OK(db2->Put(wo, Key(2), DummyString(20000)));
wait_flush();
ASSERT_OK(Put(2, Key(1), DummyString(1), wo));
wait_flush();
ASSERT_OK(static_cast<DBImpl*>(db2)->TEST_WaitForFlushMemTable());
{
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default") +
GetNumberOfSstFilesForColumnFamily(db_, "cf1") +
GetNumberOfSstFilesForColumnFamily(db_, "cf2"),
static_cast<uint64_t>(1));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db2, "default"),
static_cast<uint64_t>(0));
}
// Triggering to flush another CF in DB1
ASSERT_OK(db2->Put(wo, Key(2), DummyString(70000)));
wait_flush();
ASSERT_OK(Put(2, Key(1), DummyString(1), wo));
wait_flush();
{
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
static_cast<uint64_t>(1));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "cf1"),
static_cast<uint64_t>(0));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "cf2"),
static_cast<uint64_t>(1));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db2, "default"),
static_cast<uint64_t>(0));
}
// Triggering flush in DB2.
ASSERT_OK(db2->Put(wo, Key(3), DummyString(40000)));
wait_flush();
ASSERT_OK(db2->Put(wo, Key(1), DummyString(1)));
wait_flush();
ASSERT_OK(static_cast<DBImpl*>(db2)->TEST_WaitForFlushMemTable());
{
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "default"),
static_cast<uint64_t>(1));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "cf1"),
static_cast<uint64_t>(0));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db_, "cf2"),
static_cast<uint64_t>(1));
ASSERT_EQ(GetNumberOfSstFilesForColumnFamily(db2, "default"),
static_cast<uint64_t>(1));
}
delete db2;
ASSERT_OK(DestroyDB(dbname2, options));
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, TestWriteBufferNoLimitWithCache) {
Options options = CurrentOptions();
options.arena_block_size = 4096;
std::shared_ptr<Cache> cache = NewLRUCache(LRUCacheOptions(
10000000 /* capacity */, 1 /* num_shard_bits */,
false /* strict_capacity_limit */, 0.0 /* high_pri_pool_ratio */,
nullptr /* memory_allocator */, kDefaultToAdaptiveMutex,
kDontChargeCacheMetadata));
options.write_buffer_size = 50000; // this is never hit
// Use a write buffer total size so that the soft limit is about
// 105000.
options.write_buffer_manager.reset(new WriteBufferManager(0, cache));
Reopen(options);
ASSERT_OK(Put("foo", "bar"));
// One dummy entry is 256KB.
ASSERT_GT(cache->GetUsage(), 128000);
}
namespace {
void ValidateKeyExistence(DB* db, const std::vector<Slice>& keys_must_exist,
const std::vector<Slice>& keys_must_not_exist) {
// Ensure that expected keys exist
std::vector<std::string> values;
if (keys_must_exist.size() > 0) {
std::vector<Status> status_list =
db->MultiGet(ReadOptions(), keys_must_exist, &values);
for (size_t i = 0; i < keys_must_exist.size(); i++) {
ASSERT_OK(status_list[i]);
}
}
// Ensure that given keys don't exist
if (keys_must_not_exist.size() > 0) {
std::vector<Status> status_list =
db->MultiGet(ReadOptions(), keys_must_not_exist, &values);
for (size_t i = 0; i < keys_must_not_exist.size(); i++) {
ASSERT_TRUE(status_list[i].IsNotFound());
}
}
}
} // namespace
TEST_F(DBTest2, WalFilterTest) {
class TestWalFilter : public WalFilter {
private:
// Processing option that is requested to be applied at the given index
WalFilter::WalProcessingOption wal_processing_option_;
// Index at which to apply wal_processing_option_
// At other indexes default wal_processing_option::kContinueProcessing is
// returned.
size_t apply_option_at_record_index_;
// Current record index, incremented with each record encountered.
size_t current_record_index_;
public:
TestWalFilter(WalFilter::WalProcessingOption wal_processing_option,
size_t apply_option_for_record_index)
: wal_processing_option_(wal_processing_option),
apply_option_at_record_index_(apply_option_for_record_index),
current_record_index_(0) {}
WalProcessingOption LogRecord(const WriteBatch& /*batch*/,
WriteBatch* /*new_batch*/,
bool* /*batch_changed*/) const override {
WalFilter::WalProcessingOption option_to_return;
if (current_record_index_ == apply_option_at_record_index_) {
option_to_return = wal_processing_option_;
}
else {
option_to_return = WalProcessingOption::kContinueProcessing;
}
// Filter is passed as a const object for RocksDB to not modify the
// object, however we modify it for our own purpose here and hence
// cast the constness away.
(const_cast<TestWalFilter*>(this)->current_record_index_)++;
return option_to_return;
}
const char* Name() const override { return "TestWalFilter"; }
};
// Create 3 batches with two keys each
std::vector<std::vector<std::string>> batch_keys(3);
batch_keys[0].push_back("key1");
batch_keys[0].push_back("key2");
batch_keys[1].push_back("key3");
batch_keys[1].push_back("key4");
batch_keys[2].push_back("key5");
batch_keys[2].push_back("key6");
// Test with all WAL processing options
for (int option = 0;
option < static_cast<int>(
WalFilter::WalProcessingOption::kWalProcessingOptionMax);
option++) {
Options options = OptionsForLogIterTest();
DestroyAndReopen(options);
CreateAndReopenWithCF({ "pikachu" }, options);
// Write given keys in given batches
for (size_t i = 0; i < batch_keys.size(); i++) {
WriteBatch batch;
for (size_t j = 0; j < batch_keys[i].size(); j++) {
ASSERT_OK(batch.Put(handles_[0], batch_keys[i][j], DummyString(1024)));
}
ASSERT_OK(dbfull()->Write(WriteOptions(), &batch));
}
WalFilter::WalProcessingOption wal_processing_option =
static_cast<WalFilter::WalProcessingOption>(option);
// Create a test filter that would apply wal_processing_option at the first
// record
size_t apply_option_for_record_index = 1;
TestWalFilter test_wal_filter(wal_processing_option,
apply_option_for_record_index);
// Reopen database with option to use WAL filter
options = OptionsForLogIterTest();
options.wal_filter = &test_wal_filter;
Status status =
TryReopenWithColumnFamilies({ "default", "pikachu" }, options);
if (wal_processing_option ==
WalFilter::WalProcessingOption::kCorruptedRecord) {
ASSERT_NOK(status);
// In case of corruption we can turn off paranoid_checks to reopen
// databse
options.paranoid_checks = false;
ReopenWithColumnFamilies({ "default", "pikachu" }, options);
}
else {
ASSERT_OK(status);
}
// Compute which keys we expect to be found
// and which we expect not to be found after recovery.
std::vector<Slice> keys_must_exist;
std::vector<Slice> keys_must_not_exist;
switch (wal_processing_option) {
case WalFilter::WalProcessingOption::kCorruptedRecord:
case WalFilter::WalProcessingOption::kContinueProcessing: {
fprintf(stderr, "Testing with complete WAL processing\n");
// we expect all records to be processed
for (size_t i = 0; i < batch_keys.size(); i++) {
for (size_t j = 0; j < batch_keys[i].size(); j++) {
keys_must_exist.push_back(Slice(batch_keys[i][j]));
}
}
break;
}
case WalFilter::WalProcessingOption::kIgnoreCurrentRecord: {
fprintf(stderr,
"Testing with ignoring record %" ROCKSDB_PRIszt " only\n",
apply_option_for_record_index);
// We expect the record with apply_option_for_record_index to be not
// found.
for (size_t i = 0; i < batch_keys.size(); i++) {
for (size_t j = 0; j < batch_keys[i].size(); j++) {
if (i == apply_option_for_record_index) {
keys_must_not_exist.push_back(Slice(batch_keys[i][j]));
}
else {
keys_must_exist.push_back(Slice(batch_keys[i][j]));
}
}
}
break;
}
case WalFilter::WalProcessingOption::kStopReplay: {
fprintf(stderr,
"Testing with stopping replay from record %" ROCKSDB_PRIszt
"\n",
apply_option_for_record_index);
// We expect records beyond apply_option_for_record_index to be not
// found.
for (size_t i = 0; i < batch_keys.size(); i++) {
for (size_t j = 0; j < batch_keys[i].size(); j++) {
if (i >= apply_option_for_record_index) {
keys_must_not_exist.push_back(Slice(batch_keys[i][j]));
}
else {
keys_must_exist.push_back(Slice(batch_keys[i][j]));
}
}
}
break;
}
default:
FAIL(); // unhandled case
}
bool checked_after_reopen = false;
while (true) {
// Ensure that expected keys exists
// and not expected keys don't exist after recovery
ValidateKeyExistence(db_, keys_must_exist, keys_must_not_exist);
if (checked_after_reopen) {
break;
}
// reopen database again to make sure previous log(s) are not used
//(even if they were skipped)
// reopn database with option to use WAL filter
options = OptionsForLogIterTest();
ReopenWithColumnFamilies({ "default", "pikachu" }, options);
checked_after_reopen = true;
}
}
}
TEST_F(DBTest2, WalFilterTestWithChangeBatch) {
class ChangeBatchHandler : public WriteBatch::Handler {
private:
// Batch to insert keys in
WriteBatch* new_write_batch_;
// Number of keys to add in the new batch
size_t num_keys_to_add_in_new_batch_;
// Number of keys added to new batch
size_t num_keys_added_;
public:
ChangeBatchHandler(WriteBatch* new_write_batch,
size_t num_keys_to_add_in_new_batch)
: new_write_batch_(new_write_batch),
num_keys_to_add_in_new_batch_(num_keys_to_add_in_new_batch),
num_keys_added_(0) {}
void Put(const Slice& key, const Slice& value) override {
if (num_keys_added_ < num_keys_to_add_in_new_batch_) {
ASSERT_OK(new_write_batch_->Put(key, value));
++num_keys_added_;
}
}
};
class TestWalFilterWithChangeBatch : public WalFilter {
private:
// Index at which to start changing records
size_t change_records_from_index_;
// Number of keys to add in the new batch
size_t num_keys_to_add_in_new_batch_;
// Current record index, incremented with each record encountered.
size_t current_record_index_;
public:
TestWalFilterWithChangeBatch(size_t change_records_from_index,
size_t num_keys_to_add_in_new_batch)
: change_records_from_index_(change_records_from_index),
num_keys_to_add_in_new_batch_(num_keys_to_add_in_new_batch),
current_record_index_(0) {}
WalProcessingOption LogRecord(const WriteBatch& batch,
WriteBatch* new_batch,
bool* batch_changed) const override {
if (current_record_index_ >= change_records_from_index_) {
ChangeBatchHandler handler(new_batch, num_keys_to_add_in_new_batch_);
Status s = batch.Iterate(&handler);
if (s.ok()) {
*batch_changed = true;
} else {
assert(false);
}
}
// Filter is passed as a const object for RocksDB to not modify the
// object, however we modify it for our own purpose here and hence
// cast the constness away.
(const_cast<TestWalFilterWithChangeBatch*>(this)
->current_record_index_)++;
return WalProcessingOption::kContinueProcessing;
}
const char* Name() const override { return "TestWalFilterWithChangeBatch"; }
};
std::vector<std::vector<std::string>> batch_keys(3);
batch_keys[0].push_back("key1");
batch_keys[0].push_back("key2");
batch_keys[1].push_back("key3");
batch_keys[1].push_back("key4");
batch_keys[2].push_back("key5");
batch_keys[2].push_back("key6");
Options options = OptionsForLogIterTest();
DestroyAndReopen(options);
CreateAndReopenWithCF({ "pikachu" }, options);
// Write given keys in given batches
for (size_t i = 0; i < batch_keys.size(); i++) {
WriteBatch batch;
for (size_t j = 0; j < batch_keys[i].size(); j++) {
ASSERT_OK(batch.Put(handles_[0], batch_keys[i][j], DummyString(1024)));
}
ASSERT_OK(dbfull()->Write(WriteOptions(), &batch));
}
// Create a test filter that would apply wal_processing_option at the first
// record
size_t change_records_from_index = 1;
size_t num_keys_to_add_in_new_batch = 1;
TestWalFilterWithChangeBatch test_wal_filter_with_change_batch(
change_records_from_index, num_keys_to_add_in_new_batch);
// Reopen database with option to use WAL filter
options = OptionsForLogIterTest();
options.wal_filter = &test_wal_filter_with_change_batch;
ReopenWithColumnFamilies({ "default", "pikachu" }, options);
// Ensure that all keys exist before change_records_from_index_
// And after that index only single key exists
// as our filter adds only single key for each batch
std::vector<Slice> keys_must_exist;
std::vector<Slice> keys_must_not_exist;
for (size_t i = 0; i < batch_keys.size(); i++) {
for (size_t j = 0; j < batch_keys[i].size(); j++) {
if (i >= change_records_from_index && j >= num_keys_to_add_in_new_batch) {
keys_must_not_exist.push_back(Slice(batch_keys[i][j]));
}
else {
keys_must_exist.push_back(Slice(batch_keys[i][j]));
}
}
}
bool checked_after_reopen = false;
while (true) {
// Ensure that expected keys exists
// and not expected keys don't exist after recovery
ValidateKeyExistence(db_, keys_must_exist, keys_must_not_exist);
if (checked_after_reopen) {
break;
}
// reopen database again to make sure previous log(s) are not used
//(even if they were skipped)
// reopn database with option to use WAL filter
options = OptionsForLogIterTest();
ReopenWithColumnFamilies({ "default", "pikachu" }, options);
checked_after_reopen = true;
}
}
TEST_F(DBTest2, WalFilterTestWithChangeBatchExtraKeys) {
class TestWalFilterWithChangeBatchAddExtraKeys : public WalFilter {
public:
WalProcessingOption LogRecord(const WriteBatch& batch, WriteBatch* new_batch,
bool* batch_changed) const override {
*new_batch = batch;
Status s = new_batch->Put("key_extra", "value_extra");
if (s.ok()) {
*batch_changed = true;
} else {
assert(false);
}
return WalProcessingOption::kContinueProcessing;
}
const char* Name() const override {
return "WalFilterTestWithChangeBatchExtraKeys";
}
};
std::vector<std::vector<std::string>> batch_keys(3);
batch_keys[0].push_back("key1");
batch_keys[0].push_back("key2");
batch_keys[1].push_back("key3");
batch_keys[1].push_back("key4");
batch_keys[2].push_back("key5");
batch_keys[2].push_back("key6");
Options options = OptionsForLogIterTest();
DestroyAndReopen(options);
CreateAndReopenWithCF({ "pikachu" }, options);
// Write given keys in given batches
for (size_t i = 0; i < batch_keys.size(); i++) {
WriteBatch batch;
for (size_t j = 0; j < batch_keys[i].size(); j++) {
ASSERT_OK(batch.Put(handles_[0], batch_keys[i][j], DummyString(1024)));
}
ASSERT_OK(dbfull()->Write(WriteOptions(), &batch));
}
// Create a test filter that would add extra keys
TestWalFilterWithChangeBatchAddExtraKeys test_wal_filter_extra_keys;
// Reopen database with option to use WAL filter
options = OptionsForLogIterTest();
options.wal_filter = &test_wal_filter_extra_keys;
Status status = TryReopenWithColumnFamilies({"default", "pikachu"}, options);
ASSERT_TRUE(status.IsNotSupported());
// Reopen without filter, now reopen should succeed - previous
// attempt to open must not have altered the db.
options = OptionsForLogIterTest();
ReopenWithColumnFamilies({ "default", "pikachu" }, options);
std::vector<Slice> keys_must_exist;
std::vector<Slice> keys_must_not_exist; // empty vector
for (size_t i = 0; i < batch_keys.size(); i++) {
for (size_t j = 0; j < batch_keys[i].size(); j++) {
keys_must_exist.push_back(Slice(batch_keys[i][j]));
}
}
ValidateKeyExistence(db_, keys_must_exist, keys_must_not_exist);
}
TEST_F(DBTest2, WalFilterTestWithColumnFamilies) {
class TestWalFilterWithColumnFamilies : public WalFilter {
private:
// column_family_id -> log_number map (provided to WALFilter)
std::map<uint32_t, uint64_t> cf_log_number_map_;
// column_family_name -> column_family_id map (provided to WALFilter)
std::map<std::string, uint32_t> cf_name_id_map_;
// column_family_name -> keys_found_in_wal map
// We store keys that are applicable to the column_family
// during recovery (i.e. aren't already flushed to SST file(s))
// for verification against the keys we expect.
std::map<uint32_t, std::vector<std::string>> cf_wal_keys_;
public:
void ColumnFamilyLogNumberMap(
const std::map<uint32_t, uint64_t>& cf_lognumber_map,
const std::map<std::string, uint32_t>& cf_name_id_map) override {
cf_log_number_map_ = cf_lognumber_map;
cf_name_id_map_ = cf_name_id_map;
}
WalProcessingOption LogRecordFound(unsigned long long log_number,
const std::string& /*log_file_name*/,
const WriteBatch& batch,
WriteBatch* /*new_batch*/,
bool* /*batch_changed*/) override {
class LogRecordBatchHandler : public WriteBatch::Handler {
private:
const std::map<uint32_t, uint64_t> & cf_log_number_map_;
std::map<uint32_t, std::vector<std::string>> & cf_wal_keys_;
unsigned long long log_number_;
public:
LogRecordBatchHandler(unsigned long long current_log_number,
const std::map<uint32_t, uint64_t> & cf_log_number_map,
std::map<uint32_t, std::vector<std::string>> & cf_wal_keys) :
cf_log_number_map_(cf_log_number_map),
cf_wal_keys_(cf_wal_keys),
log_number_(current_log_number){}
Status PutCF(uint32_t column_family_id, const Slice& key,
const Slice& /*value*/) override {
auto it = cf_log_number_map_.find(column_family_id);
assert(it != cf_log_number_map_.end());
unsigned long long log_number_for_cf = it->second;
// If the current record is applicable for column_family_id
// (i.e. isn't flushed to SST file(s) for column_family_id)
// add it to the cf_wal_keys_ map for verification.
if (log_number_ >= log_number_for_cf) {
cf_wal_keys_[column_family_id].push_back(std::string(key.data(),
key.size()));
}
return Status::OK();
}
} handler(log_number, cf_log_number_map_, cf_wal_keys_);
Status s = batch.Iterate(&handler);
if (!s.ok()) {
// TODO(AR) is this ok?
return WalProcessingOption::kCorruptedRecord;
}
return WalProcessingOption::kContinueProcessing;
}
const char* Name() const override {
return "WalFilterTestWithColumnFamilies";
}
const std::map<uint32_t, std::vector<std::string>>& GetColumnFamilyKeys() {
return cf_wal_keys_;
}
const std::map<std::string, uint32_t> & GetColumnFamilyNameIdMap() {
return cf_name_id_map_;
}
};
std::vector<std::vector<std::string>> batch_keys_pre_flush(3);
batch_keys_pre_flush[0].push_back("key1");
batch_keys_pre_flush[0].push_back("key2");
batch_keys_pre_flush[1].push_back("key3");
batch_keys_pre_flush[1].push_back("key4");
batch_keys_pre_flush[2].push_back("key5");
batch_keys_pre_flush[2].push_back("key6");
Options options = OptionsForLogIterTest();
DestroyAndReopen(options);
CreateAndReopenWithCF({ "pikachu" }, options);
// Write given keys in given batches
for (size_t i = 0; i < batch_keys_pre_flush.size(); i++) {
WriteBatch batch;
for (size_t j = 0; j < batch_keys_pre_flush[i].size(); j++) {
ASSERT_OK(batch.Put(handles_[0], batch_keys_pre_flush[i][j],
DummyString(1024)));
ASSERT_OK(batch.Put(handles_[1], batch_keys_pre_flush[i][j],
DummyString(1024)));
}
ASSERT_OK(dbfull()->Write(WriteOptions(), &batch));
}
//Flush default column-family
ASSERT_OK(db_->Flush(FlushOptions(), handles_[0]));
// Do some more writes
std::vector<std::vector<std::string>> batch_keys_post_flush(3);
batch_keys_post_flush[0].push_back("key7");
batch_keys_post_flush[0].push_back("key8");
batch_keys_post_flush[1].push_back("key9");
batch_keys_post_flush[1].push_back("key10");
batch_keys_post_flush[2].push_back("key11");
batch_keys_post_flush[2].push_back("key12");
// Write given keys in given batches
for (size_t i = 0; i < batch_keys_post_flush.size(); i++) {
WriteBatch batch;
for (size_t j = 0; j < batch_keys_post_flush[i].size(); j++) {
ASSERT_OK(batch.Put(handles_[0], batch_keys_post_flush[i][j],
DummyString(1024)));
ASSERT_OK(batch.Put(handles_[1], batch_keys_post_flush[i][j],
DummyString(1024)));
}
ASSERT_OK(dbfull()->Write(WriteOptions(), &batch));
}
// On Recovery we should only find the second batch applicable to default CF
// But both batches applicable to pikachu CF
// Create a test filter that would add extra keys
TestWalFilterWithColumnFamilies test_wal_filter_column_families;
// Reopen database with option to use WAL filter
options = OptionsForLogIterTest();
options.wal_filter = &test_wal_filter_column_families;
Status status =
TryReopenWithColumnFamilies({ "default", "pikachu" }, options);
ASSERT_TRUE(status.ok());
// verify that handles_[0] only has post_flush keys
// while handles_[1] has pre and post flush keys
auto cf_wal_keys = test_wal_filter_column_families.GetColumnFamilyKeys();
auto name_id_map = test_wal_filter_column_families.GetColumnFamilyNameIdMap();
size_t index = 0;
auto keys_cf = cf_wal_keys[name_id_map[kDefaultColumnFamilyName]];
//default column-family, only post_flush keys are expected
for (size_t i = 0; i < batch_keys_post_flush.size(); i++) {
for (size_t j = 0; j < batch_keys_post_flush[i].size(); j++) {
Slice key_from_the_log(keys_cf[index++]);
Slice batch_key(batch_keys_post_flush[i][j]);
ASSERT_EQ(key_from_the_log.compare(batch_key), 0);
}
}
ASSERT_EQ(index, keys_cf.size());
index = 0;
keys_cf = cf_wal_keys[name_id_map["pikachu"]];
//pikachu column-family, all keys are expected
for (size_t i = 0; i < batch_keys_pre_flush.size(); i++) {
for (size_t j = 0; j < batch_keys_pre_flush[i].size(); j++) {
Slice key_from_the_log(keys_cf[index++]);
Slice batch_key(batch_keys_pre_flush[i][j]);
ASSERT_EQ(key_from_the_log.compare(batch_key), 0);
}
}
for (size_t i = 0; i < batch_keys_post_flush.size(); i++) {
for (size_t j = 0; j < batch_keys_post_flush[i].size(); j++) {
Slice key_from_the_log(keys_cf[index++]);
Slice batch_key(batch_keys_post_flush[i][j]);
ASSERT_EQ(key_from_the_log.compare(batch_key), 0);
}
}
ASSERT_EQ(index, keys_cf.size());
}
TEST_F(DBTest2, PresetCompressionDict) {
Reduce scope of compression dictionary to single SST (#4952) Summary: Our previous approach was to train one compression dictionary per compaction, using the first output SST to train a dictionary, and then applying it on subsequent SSTs in the same compaction. While this was great for minimizing CPU/memory/I/O overhead, it did not achieve good compression ratios in practice. In our most promising potential use case, moderate reductions in a dictionary's scope make a major difference on compression ratio. So, this PR changes compression dictionary to be scoped per-SST. It accepts the tradeoff during table building to use more memory and CPU. Important changes include: - The `BlockBasedTableBuilder` has a new state when dictionary compression is in-use: `kBuffered`. In that state it accumulates uncompressed data in-memory whenever `Add` is called. - After accumulating target file size bytes or calling `BlockBasedTableBuilder::Finish`, a `BlockBasedTableBuilder` moves to the `kUnbuffered` state. The transition (`EnterUnbuffered()`) involves sampling the buffered data, training a dictionary, and compressing/writing out all buffered data. In the `kUnbuffered` state, a `BlockBasedTableBuilder` behaves the same as before -- blocks are compressed/written out as soon as they fill up. - Samples are now whole uncompressed data blocks, except the final sample may be a partial data block so we don't breach the user's configured `max_dict_bytes` or `zstd_max_train_bytes`. The dictionary trainer is supposed to work better when we pass it real units of compression. Previously we were passing 64-byte KV samples which was not realistic. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4952 Differential Revision: D13967980 Pulled By: ajkr fbshipit-source-id: 82bea6f7537e1529c7a1a4cdee84585f5949300f
6 years ago
// Verifies that compression ratio improves when dictionary is enabled, and
// improves even further when the dictionary is trained by ZSTD.
const size_t kBlockSizeBytes = 4 << 10;
const size_t kL0FileBytes = 128 << 10;
const size_t kApproxPerBlockOverheadBytes = 50;
const int kNumL0Files = 5;
Options options;
// Make sure to use any custom env that the test is configured with.
options.env = CurrentOptions().env;
options.allow_concurrent_memtable_write = false;
options.arena_block_size = kBlockSizeBytes;
options.create_if_missing = true;
options.disable_auto_compactions = true;
options.level0_file_num_compaction_trigger = kNumL0Files;
options.memtable_factory.reset(
test::NewSpecialSkipListFactory(kL0FileBytes / kBlockSizeBytes));
options.num_levels = 2;
options.target_file_size_base = kL0FileBytes;
options.target_file_size_multiplier = 2;
options.write_buffer_size = kL0FileBytes;
BlockBasedTableOptions table_options;
table_options.block_size = kBlockSizeBytes;
std::vector<CompressionType> compression_types;
if (Zlib_Supported()) {
compression_types.push_back(kZlibCompression);
}
#if LZ4_VERSION_NUMBER >= 10400 // r124+
compression_types.push_back(kLZ4Compression);
compression_types.push_back(kLZ4HCCompression);
#endif // LZ4_VERSION_NUMBER >= 10400
if (ZSTD_Supported()) {
compression_types.push_back(kZSTD);
}
enum DictionaryTypes : int {
kWithoutDict,
kWithDict,
Support using ZDICT_finalizeDictionary to generate zstd dictionary (#9857) Summary: An untrained dictionary is currently simply the concatenation of several samples. The ZSTD API, ZDICT_finalizeDictionary(), can improve such a dictionary's effectiveness at low cost. This PR changes how dictionary is created by calling the ZSTD ZDICT_finalizeDictionary() API instead of creating raw content dictionary (when max_dict_buffer_bytes > 0), and pass in all buffered uncompressed data blocks as samples. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9857 Test Plan: #### db_bench test for cpu/memory of compression+decompression and space saving on synthetic data: Set up: change the parameter [here](https://github.com/facebook/rocksdb/blob/fb9a167a55e0970b1ef6f67c1600c8d9c4c6114f/tools/db_bench_tool.cc#L1766) to 16384 to make synthetic data more compressible. ``` # linked local ZSTD with version 1.5.2 # DEBUG_LEVEL=0 ROCKSDB_NO_FBCODE=1 ROCKSDB_DISABLE_ZSTD=1 EXTRA_CXXFLAGS="-DZSTD_STATIC_LINKING_ONLY -DZSTD -I/data/users/changyubi/install/include/" EXTRA_LDFLAGS="-L/data/users/changyubi/install/lib/ -l:libzstd.a" make -j32 db_bench dict_bytes=16384 train_bytes=1048576 echo "========== No Dictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=0 -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=0 -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== Raw Content Dictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench_main -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench_main -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== FinalizeDictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== TrainDictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total # Result: TrainDictionary is much better on space saving, but FinalizeDictionary seems to use less memory. # before compression data size: 1.2GB dict_bytes=16384 max_dict_buffer_bytes = 1048576 space cpu/memory No Dictionary 468M 14.93user 1.00system 0:15.92elapsed 100%CPU (0avgtext+0avgdata 23904maxresident)k Raw Dictionary 251M 15.81user 0.80system 0:16.56elapsed 100%CPU (0avgtext+0avgdata 156808maxresident)k FinalizeDictionary 236M 11.93user 0.64system 0:12.56elapsed 100%CPU (0avgtext+0avgdata 89548maxresident)k TrainDictionary 84M 7.29user 0.45system 0:07.75elapsed 100%CPU (0avgtext+0avgdata 97288maxresident)k ``` #### Benchmark on 10 sample SST files for spacing saving and CPU time on compression: FinalizeDictionary is comparable to TrainDictionary in terms of space saving, and takes less time in compression. ``` dict_bytes=16384 train_bytes=1048576 for sst_file in `ls ../temp/myrock-sst/` do echo "********** $sst_file **********" echo "========== No Dictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD echo "========== Raw Content Dictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes echo "========== FinalizeDictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes --compression_zstd_max_train_bytes=$train_bytes --compression_use_zstd_finalize_dict echo "========== TrainDictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes --compression_zstd_max_train_bytes=$train_bytes done 010240.sst (Size/Time) 011029.sst 013184.sst 021552.sst 185054.sst 185137.sst 191666.sst 7560381.sst 7604174.sst 7635312.sst No Dictionary 28165569 / 2614419 32899411 / 2976832 32977848 / 3055542 31966329 / 2004590 33614351 / 1755877 33429029 / 1717042 33611933 / 1776936 33634045 / 2771417 33789721 / 2205414 33592194 / 388254 Raw Content Dictionary 28019950 / 2697961 33748665 / 3572422 33896373 / 3534701 26418431 / 2259658 28560825 / 1839168 28455030 / 1846039 28494319 / 1861349 32391599 / 3095649 33772142 / 2407843 33592230 / 474523 FinalizeDictionary 27896012 / 2650029 33763886 / 3719427 33904283 / 3552793 26008225 / 2198033 28111872 / 1869530 28014374 / 1789771 28047706 / 1848300 32296254 / 3204027 33698698 / 2381468 33592344 / 517433 TrainDictionary 28046089 / 2740037 33706480 / 3679019 33885741 / 3629351 25087123 / 2204558 27194353 / 1970207 27234229 / 1896811 27166710 / 1903119 32011041 / 3322315 32730692 / 2406146 33608631 / 570593 ``` #### Decompression/Read test: With FinalizeDictionary/TrainDictionary, some data structure used for decompression are in stored in dictionary, so they are expected to be faster in terms of decompression/reads. ``` dict_bytes=16384 train_bytes=1048576 echo "No Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=0 > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=0 2>&1 | grep MB/s echo "Raw Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes 2>&1 | grep MB/s echo "FinalizeDict" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false 2>&1 | grep MB/s echo "Train Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes 2>&1 | grep MB/s No Dictionary readrandom : 12.183 micros/op 82082 ops/sec 12.183 seconds 1000000 operations; 9.1 MB/s (1000000 of 1000000 found) Raw Dictionary readrandom : 12.314 micros/op 81205 ops/sec 12.314 seconds 1000000 operations; 9.0 MB/s (1000000 of 1000000 found) FinalizeDict readrandom : 9.787 micros/op 102180 ops/sec 9.787 seconds 1000000 operations; 11.3 MB/s (1000000 of 1000000 found) Train Dictionary readrandom : 9.698 micros/op 103108 ops/sec 9.699 seconds 1000000 operations; 11.4 MB/s (1000000 of 1000000 found) ``` Reviewed By: ajkr Differential Revision: D35720026 Pulled By: cbi42 fbshipit-source-id: 24d230fdff0fd28a1bb650658798f00dfcfb2a1f
2 years ago
kWithZSTDfinalizeDict,
kWithZSTDTrainedDict,
kDictEnd,
};
for (auto compression_type : compression_types) {
options.compression = compression_type;
size_t bytes_without_dict = 0;
size_t bytes_with_dict = 0;
Support using ZDICT_finalizeDictionary to generate zstd dictionary (#9857) Summary: An untrained dictionary is currently simply the concatenation of several samples. The ZSTD API, ZDICT_finalizeDictionary(), can improve such a dictionary's effectiveness at low cost. This PR changes how dictionary is created by calling the ZSTD ZDICT_finalizeDictionary() API instead of creating raw content dictionary (when max_dict_buffer_bytes > 0), and pass in all buffered uncompressed data blocks as samples. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9857 Test Plan: #### db_bench test for cpu/memory of compression+decompression and space saving on synthetic data: Set up: change the parameter [here](https://github.com/facebook/rocksdb/blob/fb9a167a55e0970b1ef6f67c1600c8d9c4c6114f/tools/db_bench_tool.cc#L1766) to 16384 to make synthetic data more compressible. ``` # linked local ZSTD with version 1.5.2 # DEBUG_LEVEL=0 ROCKSDB_NO_FBCODE=1 ROCKSDB_DISABLE_ZSTD=1 EXTRA_CXXFLAGS="-DZSTD_STATIC_LINKING_ONLY -DZSTD -I/data/users/changyubi/install/include/" EXTRA_LDFLAGS="-L/data/users/changyubi/install/lib/ -l:libzstd.a" make -j32 db_bench dict_bytes=16384 train_bytes=1048576 echo "========== No Dictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=0 -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=0 -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== Raw Content Dictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench_main -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench_main -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== FinalizeDictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== TrainDictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total # Result: TrainDictionary is much better on space saving, but FinalizeDictionary seems to use less memory. # before compression data size: 1.2GB dict_bytes=16384 max_dict_buffer_bytes = 1048576 space cpu/memory No Dictionary 468M 14.93user 1.00system 0:15.92elapsed 100%CPU (0avgtext+0avgdata 23904maxresident)k Raw Dictionary 251M 15.81user 0.80system 0:16.56elapsed 100%CPU (0avgtext+0avgdata 156808maxresident)k FinalizeDictionary 236M 11.93user 0.64system 0:12.56elapsed 100%CPU (0avgtext+0avgdata 89548maxresident)k TrainDictionary 84M 7.29user 0.45system 0:07.75elapsed 100%CPU (0avgtext+0avgdata 97288maxresident)k ``` #### Benchmark on 10 sample SST files for spacing saving and CPU time on compression: FinalizeDictionary is comparable to TrainDictionary in terms of space saving, and takes less time in compression. ``` dict_bytes=16384 train_bytes=1048576 for sst_file in `ls ../temp/myrock-sst/` do echo "********** $sst_file **********" echo "========== No Dictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD echo "========== Raw Content Dictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes echo "========== FinalizeDictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes --compression_zstd_max_train_bytes=$train_bytes --compression_use_zstd_finalize_dict echo "========== TrainDictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes --compression_zstd_max_train_bytes=$train_bytes done 010240.sst (Size/Time) 011029.sst 013184.sst 021552.sst 185054.sst 185137.sst 191666.sst 7560381.sst 7604174.sst 7635312.sst No Dictionary 28165569 / 2614419 32899411 / 2976832 32977848 / 3055542 31966329 / 2004590 33614351 / 1755877 33429029 / 1717042 33611933 / 1776936 33634045 / 2771417 33789721 / 2205414 33592194 / 388254 Raw Content Dictionary 28019950 / 2697961 33748665 / 3572422 33896373 / 3534701 26418431 / 2259658 28560825 / 1839168 28455030 / 1846039 28494319 / 1861349 32391599 / 3095649 33772142 / 2407843 33592230 / 474523 FinalizeDictionary 27896012 / 2650029 33763886 / 3719427 33904283 / 3552793 26008225 / 2198033 28111872 / 1869530 28014374 / 1789771 28047706 / 1848300 32296254 / 3204027 33698698 / 2381468 33592344 / 517433 TrainDictionary 28046089 / 2740037 33706480 / 3679019 33885741 / 3629351 25087123 / 2204558 27194353 / 1970207 27234229 / 1896811 27166710 / 1903119 32011041 / 3322315 32730692 / 2406146 33608631 / 570593 ``` #### Decompression/Read test: With FinalizeDictionary/TrainDictionary, some data structure used for decompression are in stored in dictionary, so they are expected to be faster in terms of decompression/reads. ``` dict_bytes=16384 train_bytes=1048576 echo "No Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=0 > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=0 2>&1 | grep MB/s echo "Raw Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes 2>&1 | grep MB/s echo "FinalizeDict" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false 2>&1 | grep MB/s echo "Train Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes 2>&1 | grep MB/s No Dictionary readrandom : 12.183 micros/op 82082 ops/sec 12.183 seconds 1000000 operations; 9.1 MB/s (1000000 of 1000000 found) Raw Dictionary readrandom : 12.314 micros/op 81205 ops/sec 12.314 seconds 1000000 operations; 9.0 MB/s (1000000 of 1000000 found) FinalizeDict readrandom : 9.787 micros/op 102180 ops/sec 9.787 seconds 1000000 operations; 11.3 MB/s (1000000 of 1000000 found) Train Dictionary readrandom : 9.698 micros/op 103108 ops/sec 9.699 seconds 1000000 operations; 11.4 MB/s (1000000 of 1000000 found) ``` Reviewed By: ajkr Differential Revision: D35720026 Pulled By: cbi42 fbshipit-source-id: 24d230fdff0fd28a1bb650658798f00dfcfb2a1f
2 years ago
size_t bytes_with_zstd_finalize_dict = 0;
size_t bytes_with_zstd_trained_dict = 0;
for (int i = kWithoutDict; i < kDictEnd; i++) {
// First iteration: compress without preset dictionary
// Second iteration: compress with preset dictionary
// Third iteration (zstd only): compress with zstd-trained dictionary
//
// To make sure the compression dictionary has the intended effect, we
// verify the compressed size is smaller in successive iterations. Also in
// the non-first iterations, verify the data we get out is the same data
// we put in.
switch (i) {
case kWithoutDict:
options.compression_opts.max_dict_bytes = 0;
options.compression_opts.zstd_max_train_bytes = 0;
break;
case kWithDict:
options.compression_opts.max_dict_bytes = kBlockSizeBytes;
options.compression_opts.zstd_max_train_bytes = 0;
break;
Support using ZDICT_finalizeDictionary to generate zstd dictionary (#9857) Summary: An untrained dictionary is currently simply the concatenation of several samples. The ZSTD API, ZDICT_finalizeDictionary(), can improve such a dictionary's effectiveness at low cost. This PR changes how dictionary is created by calling the ZSTD ZDICT_finalizeDictionary() API instead of creating raw content dictionary (when max_dict_buffer_bytes > 0), and pass in all buffered uncompressed data blocks as samples. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9857 Test Plan: #### db_bench test for cpu/memory of compression+decompression and space saving on synthetic data: Set up: change the parameter [here](https://github.com/facebook/rocksdb/blob/fb9a167a55e0970b1ef6f67c1600c8d9c4c6114f/tools/db_bench_tool.cc#L1766) to 16384 to make synthetic data more compressible. ``` # linked local ZSTD with version 1.5.2 # DEBUG_LEVEL=0 ROCKSDB_NO_FBCODE=1 ROCKSDB_DISABLE_ZSTD=1 EXTRA_CXXFLAGS="-DZSTD_STATIC_LINKING_ONLY -DZSTD -I/data/users/changyubi/install/include/" EXTRA_LDFLAGS="-L/data/users/changyubi/install/lib/ -l:libzstd.a" make -j32 db_bench dict_bytes=16384 train_bytes=1048576 echo "========== No Dictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=0 -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=0 -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== Raw Content Dictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench_main -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench_main -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== FinalizeDictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== TrainDictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total # Result: TrainDictionary is much better on space saving, but FinalizeDictionary seems to use less memory. # before compression data size: 1.2GB dict_bytes=16384 max_dict_buffer_bytes = 1048576 space cpu/memory No Dictionary 468M 14.93user 1.00system 0:15.92elapsed 100%CPU (0avgtext+0avgdata 23904maxresident)k Raw Dictionary 251M 15.81user 0.80system 0:16.56elapsed 100%CPU (0avgtext+0avgdata 156808maxresident)k FinalizeDictionary 236M 11.93user 0.64system 0:12.56elapsed 100%CPU (0avgtext+0avgdata 89548maxresident)k TrainDictionary 84M 7.29user 0.45system 0:07.75elapsed 100%CPU (0avgtext+0avgdata 97288maxresident)k ``` #### Benchmark on 10 sample SST files for spacing saving and CPU time on compression: FinalizeDictionary is comparable to TrainDictionary in terms of space saving, and takes less time in compression. ``` dict_bytes=16384 train_bytes=1048576 for sst_file in `ls ../temp/myrock-sst/` do echo "********** $sst_file **********" echo "========== No Dictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD echo "========== Raw Content Dictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes echo "========== FinalizeDictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes --compression_zstd_max_train_bytes=$train_bytes --compression_use_zstd_finalize_dict echo "========== TrainDictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes --compression_zstd_max_train_bytes=$train_bytes done 010240.sst (Size/Time) 011029.sst 013184.sst 021552.sst 185054.sst 185137.sst 191666.sst 7560381.sst 7604174.sst 7635312.sst No Dictionary 28165569 / 2614419 32899411 / 2976832 32977848 / 3055542 31966329 / 2004590 33614351 / 1755877 33429029 / 1717042 33611933 / 1776936 33634045 / 2771417 33789721 / 2205414 33592194 / 388254 Raw Content Dictionary 28019950 / 2697961 33748665 / 3572422 33896373 / 3534701 26418431 / 2259658 28560825 / 1839168 28455030 / 1846039 28494319 / 1861349 32391599 / 3095649 33772142 / 2407843 33592230 / 474523 FinalizeDictionary 27896012 / 2650029 33763886 / 3719427 33904283 / 3552793 26008225 / 2198033 28111872 / 1869530 28014374 / 1789771 28047706 / 1848300 32296254 / 3204027 33698698 / 2381468 33592344 / 517433 TrainDictionary 28046089 / 2740037 33706480 / 3679019 33885741 / 3629351 25087123 / 2204558 27194353 / 1970207 27234229 / 1896811 27166710 / 1903119 32011041 / 3322315 32730692 / 2406146 33608631 / 570593 ``` #### Decompression/Read test: With FinalizeDictionary/TrainDictionary, some data structure used for decompression are in stored in dictionary, so they are expected to be faster in terms of decompression/reads. ``` dict_bytes=16384 train_bytes=1048576 echo "No Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=0 > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=0 2>&1 | grep MB/s echo "Raw Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes 2>&1 | grep MB/s echo "FinalizeDict" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false 2>&1 | grep MB/s echo "Train Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes 2>&1 | grep MB/s No Dictionary readrandom : 12.183 micros/op 82082 ops/sec 12.183 seconds 1000000 operations; 9.1 MB/s (1000000 of 1000000 found) Raw Dictionary readrandom : 12.314 micros/op 81205 ops/sec 12.314 seconds 1000000 operations; 9.0 MB/s (1000000 of 1000000 found) FinalizeDict readrandom : 9.787 micros/op 102180 ops/sec 9.787 seconds 1000000 operations; 11.3 MB/s (1000000 of 1000000 found) Train Dictionary readrandom : 9.698 micros/op 103108 ops/sec 9.699 seconds 1000000 operations; 11.4 MB/s (1000000 of 1000000 found) ``` Reviewed By: ajkr Differential Revision: D35720026 Pulled By: cbi42 fbshipit-source-id: 24d230fdff0fd28a1bb650658798f00dfcfb2a1f
2 years ago
case kWithZSTDfinalizeDict:
if (compression_type != kZSTD ||
!ZSTD_FinalizeDictionarySupported()) {
Support using ZDICT_finalizeDictionary to generate zstd dictionary (#9857) Summary: An untrained dictionary is currently simply the concatenation of several samples. The ZSTD API, ZDICT_finalizeDictionary(), can improve such a dictionary's effectiveness at low cost. This PR changes how dictionary is created by calling the ZSTD ZDICT_finalizeDictionary() API instead of creating raw content dictionary (when max_dict_buffer_bytes > 0), and pass in all buffered uncompressed data blocks as samples. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9857 Test Plan: #### db_bench test for cpu/memory of compression+decompression and space saving on synthetic data: Set up: change the parameter [here](https://github.com/facebook/rocksdb/blob/fb9a167a55e0970b1ef6f67c1600c8d9c4c6114f/tools/db_bench_tool.cc#L1766) to 16384 to make synthetic data more compressible. ``` # linked local ZSTD with version 1.5.2 # DEBUG_LEVEL=0 ROCKSDB_NO_FBCODE=1 ROCKSDB_DISABLE_ZSTD=1 EXTRA_CXXFLAGS="-DZSTD_STATIC_LINKING_ONLY -DZSTD -I/data/users/changyubi/install/include/" EXTRA_LDFLAGS="-L/data/users/changyubi/install/lib/ -l:libzstd.a" make -j32 db_bench dict_bytes=16384 train_bytes=1048576 echo "========== No Dictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=0 -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=0 -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== Raw Content Dictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench_main -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench_main -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== FinalizeDictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== TrainDictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total # Result: TrainDictionary is much better on space saving, but FinalizeDictionary seems to use less memory. # before compression data size: 1.2GB dict_bytes=16384 max_dict_buffer_bytes = 1048576 space cpu/memory No Dictionary 468M 14.93user 1.00system 0:15.92elapsed 100%CPU (0avgtext+0avgdata 23904maxresident)k Raw Dictionary 251M 15.81user 0.80system 0:16.56elapsed 100%CPU (0avgtext+0avgdata 156808maxresident)k FinalizeDictionary 236M 11.93user 0.64system 0:12.56elapsed 100%CPU (0avgtext+0avgdata 89548maxresident)k TrainDictionary 84M 7.29user 0.45system 0:07.75elapsed 100%CPU (0avgtext+0avgdata 97288maxresident)k ``` #### Benchmark on 10 sample SST files for spacing saving and CPU time on compression: FinalizeDictionary is comparable to TrainDictionary in terms of space saving, and takes less time in compression. ``` dict_bytes=16384 train_bytes=1048576 for sst_file in `ls ../temp/myrock-sst/` do echo "********** $sst_file **********" echo "========== No Dictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD echo "========== Raw Content Dictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes echo "========== FinalizeDictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes --compression_zstd_max_train_bytes=$train_bytes --compression_use_zstd_finalize_dict echo "========== TrainDictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes --compression_zstd_max_train_bytes=$train_bytes done 010240.sst (Size/Time) 011029.sst 013184.sst 021552.sst 185054.sst 185137.sst 191666.sst 7560381.sst 7604174.sst 7635312.sst No Dictionary 28165569 / 2614419 32899411 / 2976832 32977848 / 3055542 31966329 / 2004590 33614351 / 1755877 33429029 / 1717042 33611933 / 1776936 33634045 / 2771417 33789721 / 2205414 33592194 / 388254 Raw Content Dictionary 28019950 / 2697961 33748665 / 3572422 33896373 / 3534701 26418431 / 2259658 28560825 / 1839168 28455030 / 1846039 28494319 / 1861349 32391599 / 3095649 33772142 / 2407843 33592230 / 474523 FinalizeDictionary 27896012 / 2650029 33763886 / 3719427 33904283 / 3552793 26008225 / 2198033 28111872 / 1869530 28014374 / 1789771 28047706 / 1848300 32296254 / 3204027 33698698 / 2381468 33592344 / 517433 TrainDictionary 28046089 / 2740037 33706480 / 3679019 33885741 / 3629351 25087123 / 2204558 27194353 / 1970207 27234229 / 1896811 27166710 / 1903119 32011041 / 3322315 32730692 / 2406146 33608631 / 570593 ``` #### Decompression/Read test: With FinalizeDictionary/TrainDictionary, some data structure used for decompression are in stored in dictionary, so they are expected to be faster in terms of decompression/reads. ``` dict_bytes=16384 train_bytes=1048576 echo "No Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=0 > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=0 2>&1 | grep MB/s echo "Raw Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes 2>&1 | grep MB/s echo "FinalizeDict" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false 2>&1 | grep MB/s echo "Train Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes 2>&1 | grep MB/s No Dictionary readrandom : 12.183 micros/op 82082 ops/sec 12.183 seconds 1000000 operations; 9.1 MB/s (1000000 of 1000000 found) Raw Dictionary readrandom : 12.314 micros/op 81205 ops/sec 12.314 seconds 1000000 operations; 9.0 MB/s (1000000 of 1000000 found) FinalizeDict readrandom : 9.787 micros/op 102180 ops/sec 9.787 seconds 1000000 operations; 11.3 MB/s (1000000 of 1000000 found) Train Dictionary readrandom : 9.698 micros/op 103108 ops/sec 9.699 seconds 1000000 operations; 11.4 MB/s (1000000 of 1000000 found) ``` Reviewed By: ajkr Differential Revision: D35720026 Pulled By: cbi42 fbshipit-source-id: 24d230fdff0fd28a1bb650658798f00dfcfb2a1f
2 years ago
continue;
}
options.compression_opts.max_dict_bytes = kBlockSizeBytes;
options.compression_opts.zstd_max_train_bytes = kL0FileBytes;
options.compression_opts.use_zstd_dict_trainer = false;
break;
case kWithZSTDTrainedDict:
if (compression_type != kZSTD || !ZSTD_TrainDictionarySupported()) {
continue;
}
options.compression_opts.max_dict_bytes = kBlockSizeBytes;
Reduce scope of compression dictionary to single SST (#4952) Summary: Our previous approach was to train one compression dictionary per compaction, using the first output SST to train a dictionary, and then applying it on subsequent SSTs in the same compaction. While this was great for minimizing CPU/memory/I/O overhead, it did not achieve good compression ratios in practice. In our most promising potential use case, moderate reductions in a dictionary's scope make a major difference on compression ratio. So, this PR changes compression dictionary to be scoped per-SST. It accepts the tradeoff during table building to use more memory and CPU. Important changes include: - The `BlockBasedTableBuilder` has a new state when dictionary compression is in-use: `kBuffered`. In that state it accumulates uncompressed data in-memory whenever `Add` is called. - After accumulating target file size bytes or calling `BlockBasedTableBuilder::Finish`, a `BlockBasedTableBuilder` moves to the `kUnbuffered` state. The transition (`EnterUnbuffered()`) involves sampling the buffered data, training a dictionary, and compressing/writing out all buffered data. In the `kUnbuffered` state, a `BlockBasedTableBuilder` behaves the same as before -- blocks are compressed/written out as soon as they fill up. - Samples are now whole uncompressed data blocks, except the final sample may be a partial data block so we don't breach the user's configured `max_dict_bytes` or `zstd_max_train_bytes`. The dictionary trainer is supposed to work better when we pass it real units of compression. Previously we were passing 64-byte KV samples which was not realistic. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4952 Differential Revision: D13967980 Pulled By: ajkr fbshipit-source-id: 82bea6f7537e1529c7a1a4cdee84585f5949300f
6 years ago
options.compression_opts.zstd_max_train_bytes = kL0FileBytes;
Support using ZDICT_finalizeDictionary to generate zstd dictionary (#9857) Summary: An untrained dictionary is currently simply the concatenation of several samples. The ZSTD API, ZDICT_finalizeDictionary(), can improve such a dictionary's effectiveness at low cost. This PR changes how dictionary is created by calling the ZSTD ZDICT_finalizeDictionary() API instead of creating raw content dictionary (when max_dict_buffer_bytes > 0), and pass in all buffered uncompressed data blocks as samples. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9857 Test Plan: #### db_bench test for cpu/memory of compression+decompression and space saving on synthetic data: Set up: change the parameter [here](https://github.com/facebook/rocksdb/blob/fb9a167a55e0970b1ef6f67c1600c8d9c4c6114f/tools/db_bench_tool.cc#L1766) to 16384 to make synthetic data more compressible. ``` # linked local ZSTD with version 1.5.2 # DEBUG_LEVEL=0 ROCKSDB_NO_FBCODE=1 ROCKSDB_DISABLE_ZSTD=1 EXTRA_CXXFLAGS="-DZSTD_STATIC_LINKING_ONLY -DZSTD -I/data/users/changyubi/install/include/" EXTRA_LDFLAGS="-L/data/users/changyubi/install/lib/ -l:libzstd.a" make -j32 db_bench dict_bytes=16384 train_bytes=1048576 echo "========== No Dictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=0 -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=0 -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== Raw Content Dictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench_main -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench_main -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== FinalizeDictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== TrainDictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total # Result: TrainDictionary is much better on space saving, but FinalizeDictionary seems to use less memory. # before compression data size: 1.2GB dict_bytes=16384 max_dict_buffer_bytes = 1048576 space cpu/memory No Dictionary 468M 14.93user 1.00system 0:15.92elapsed 100%CPU (0avgtext+0avgdata 23904maxresident)k Raw Dictionary 251M 15.81user 0.80system 0:16.56elapsed 100%CPU (0avgtext+0avgdata 156808maxresident)k FinalizeDictionary 236M 11.93user 0.64system 0:12.56elapsed 100%CPU (0avgtext+0avgdata 89548maxresident)k TrainDictionary 84M 7.29user 0.45system 0:07.75elapsed 100%CPU (0avgtext+0avgdata 97288maxresident)k ``` #### Benchmark on 10 sample SST files for spacing saving and CPU time on compression: FinalizeDictionary is comparable to TrainDictionary in terms of space saving, and takes less time in compression. ``` dict_bytes=16384 train_bytes=1048576 for sst_file in `ls ../temp/myrock-sst/` do echo "********** $sst_file **********" echo "========== No Dictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD echo "========== Raw Content Dictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes echo "========== FinalizeDictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes --compression_zstd_max_train_bytes=$train_bytes --compression_use_zstd_finalize_dict echo "========== TrainDictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes --compression_zstd_max_train_bytes=$train_bytes done 010240.sst (Size/Time) 011029.sst 013184.sst 021552.sst 185054.sst 185137.sst 191666.sst 7560381.sst 7604174.sst 7635312.sst No Dictionary 28165569 / 2614419 32899411 / 2976832 32977848 / 3055542 31966329 / 2004590 33614351 / 1755877 33429029 / 1717042 33611933 / 1776936 33634045 / 2771417 33789721 / 2205414 33592194 / 388254 Raw Content Dictionary 28019950 / 2697961 33748665 / 3572422 33896373 / 3534701 26418431 / 2259658 28560825 / 1839168 28455030 / 1846039 28494319 / 1861349 32391599 / 3095649 33772142 / 2407843 33592230 / 474523 FinalizeDictionary 27896012 / 2650029 33763886 / 3719427 33904283 / 3552793 26008225 / 2198033 28111872 / 1869530 28014374 / 1789771 28047706 / 1848300 32296254 / 3204027 33698698 / 2381468 33592344 / 517433 TrainDictionary 28046089 / 2740037 33706480 / 3679019 33885741 / 3629351 25087123 / 2204558 27194353 / 1970207 27234229 / 1896811 27166710 / 1903119 32011041 / 3322315 32730692 / 2406146 33608631 / 570593 ``` #### Decompression/Read test: With FinalizeDictionary/TrainDictionary, some data structure used for decompression are in stored in dictionary, so they are expected to be faster in terms of decompression/reads. ``` dict_bytes=16384 train_bytes=1048576 echo "No Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=0 > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=0 2>&1 | grep MB/s echo "Raw Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes 2>&1 | grep MB/s echo "FinalizeDict" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false 2>&1 | grep MB/s echo "Train Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes 2>&1 | grep MB/s No Dictionary readrandom : 12.183 micros/op 82082 ops/sec 12.183 seconds 1000000 operations; 9.1 MB/s (1000000 of 1000000 found) Raw Dictionary readrandom : 12.314 micros/op 81205 ops/sec 12.314 seconds 1000000 operations; 9.0 MB/s (1000000 of 1000000 found) FinalizeDict readrandom : 9.787 micros/op 102180 ops/sec 9.787 seconds 1000000 operations; 11.3 MB/s (1000000 of 1000000 found) Train Dictionary readrandom : 9.698 micros/op 103108 ops/sec 9.699 seconds 1000000 operations; 11.4 MB/s (1000000 of 1000000 found) ``` Reviewed By: ajkr Differential Revision: D35720026 Pulled By: cbi42 fbshipit-source-id: 24d230fdff0fd28a1bb650658798f00dfcfb2a1f
2 years ago
options.compression_opts.use_zstd_dict_trainer = true;
break;
default:
assert(false);
}
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
CreateAndReopenWithCF({"pikachu"}, options);
Random rnd(301);
Reduce scope of compression dictionary to single SST (#4952) Summary: Our previous approach was to train one compression dictionary per compaction, using the first output SST to train a dictionary, and then applying it on subsequent SSTs in the same compaction. While this was great for minimizing CPU/memory/I/O overhead, it did not achieve good compression ratios in practice. In our most promising potential use case, moderate reductions in a dictionary's scope make a major difference on compression ratio. So, this PR changes compression dictionary to be scoped per-SST. It accepts the tradeoff during table building to use more memory and CPU. Important changes include: - The `BlockBasedTableBuilder` has a new state when dictionary compression is in-use: `kBuffered`. In that state it accumulates uncompressed data in-memory whenever `Add` is called. - After accumulating target file size bytes or calling `BlockBasedTableBuilder::Finish`, a `BlockBasedTableBuilder` moves to the `kUnbuffered` state. The transition (`EnterUnbuffered()`) involves sampling the buffered data, training a dictionary, and compressing/writing out all buffered data. In the `kUnbuffered` state, a `BlockBasedTableBuilder` behaves the same as before -- blocks are compressed/written out as soon as they fill up. - Samples are now whole uncompressed data blocks, except the final sample may be a partial data block so we don't breach the user's configured `max_dict_bytes` or `zstd_max_train_bytes`. The dictionary trainer is supposed to work better when we pass it real units of compression. Previously we were passing 64-byte KV samples which was not realistic. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4952 Differential Revision: D13967980 Pulled By: ajkr fbshipit-source-id: 82bea6f7537e1529c7a1a4cdee84585f5949300f
6 years ago
std::string seq_datas[10];
for (int j = 0; j < 10; ++j) {
seq_datas[j] =
rnd.RandomString(kBlockSizeBytes - kApproxPerBlockOverheadBytes);
Reduce scope of compression dictionary to single SST (#4952) Summary: Our previous approach was to train one compression dictionary per compaction, using the first output SST to train a dictionary, and then applying it on subsequent SSTs in the same compaction. While this was great for minimizing CPU/memory/I/O overhead, it did not achieve good compression ratios in practice. In our most promising potential use case, moderate reductions in a dictionary's scope make a major difference on compression ratio. So, this PR changes compression dictionary to be scoped per-SST. It accepts the tradeoff during table building to use more memory and CPU. Important changes include: - The `BlockBasedTableBuilder` has a new state when dictionary compression is in-use: `kBuffered`. In that state it accumulates uncompressed data in-memory whenever `Add` is called. - After accumulating target file size bytes or calling `BlockBasedTableBuilder::Finish`, a `BlockBasedTableBuilder` moves to the `kUnbuffered` state. The transition (`EnterUnbuffered()`) involves sampling the buffered data, training a dictionary, and compressing/writing out all buffered data. In the `kUnbuffered` state, a `BlockBasedTableBuilder` behaves the same as before -- blocks are compressed/written out as soon as they fill up. - Samples are now whole uncompressed data blocks, except the final sample may be a partial data block so we don't breach the user's configured `max_dict_bytes` or `zstd_max_train_bytes`. The dictionary trainer is supposed to work better when we pass it real units of compression. Previously we were passing 64-byte KV samples which was not realistic. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4952 Differential Revision: D13967980 Pulled By: ajkr fbshipit-source-id: 82bea6f7537e1529c7a1a4cdee84585f5949300f
6 years ago
}
ASSERT_EQ(0, NumTableFilesAtLevel(0, 1));
for (int j = 0; j < kNumL0Files; ++j) {
for (size_t k = 0; k < kL0FileBytes / kBlockSizeBytes + 1; ++k) {
Reduce scope of compression dictionary to single SST (#4952) Summary: Our previous approach was to train one compression dictionary per compaction, using the first output SST to train a dictionary, and then applying it on subsequent SSTs in the same compaction. While this was great for minimizing CPU/memory/I/O overhead, it did not achieve good compression ratios in practice. In our most promising potential use case, moderate reductions in a dictionary's scope make a major difference on compression ratio. So, this PR changes compression dictionary to be scoped per-SST. It accepts the tradeoff during table building to use more memory and CPU. Important changes include: - The `BlockBasedTableBuilder` has a new state when dictionary compression is in-use: `kBuffered`. In that state it accumulates uncompressed data in-memory whenever `Add` is called. - After accumulating target file size bytes or calling `BlockBasedTableBuilder::Finish`, a `BlockBasedTableBuilder` moves to the `kUnbuffered` state. The transition (`EnterUnbuffered()`) involves sampling the buffered data, training a dictionary, and compressing/writing out all buffered data. In the `kUnbuffered` state, a `BlockBasedTableBuilder` behaves the same as before -- blocks are compressed/written out as soon as they fill up. - Samples are now whole uncompressed data blocks, except the final sample may be a partial data block so we don't breach the user's configured `max_dict_bytes` or `zstd_max_train_bytes`. The dictionary trainer is supposed to work better when we pass it real units of compression. Previously we were passing 64-byte KV samples which was not realistic. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4952 Differential Revision: D13967980 Pulled By: ajkr fbshipit-source-id: 82bea6f7537e1529c7a1a4cdee84585f5949300f
6 years ago
auto key_num = j * (kL0FileBytes / kBlockSizeBytes) + k;
ASSERT_OK(Put(1, Key(static_cast<int>(key_num)),
seq_datas[(key_num / 10) % 10]));
}
ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable(handles_[1]));
ASSERT_EQ(j + 1, NumTableFilesAtLevel(0, 1));
}
ASSERT_OK(dbfull()->TEST_CompactRange(0, nullptr, nullptr, handles_[1],
true /* disallow_trivial_move */));
ASSERT_EQ(0, NumTableFilesAtLevel(0, 1));
ASSERT_GT(NumTableFilesAtLevel(1, 1), 0);
// Get the live sst files size
size_t total_sst_bytes = TotalSize(1);
if (i == kWithoutDict) {
bytes_without_dict = total_sst_bytes;
} else if (i == kWithDict) {
bytes_with_dict = total_sst_bytes;
Support using ZDICT_finalizeDictionary to generate zstd dictionary (#9857) Summary: An untrained dictionary is currently simply the concatenation of several samples. The ZSTD API, ZDICT_finalizeDictionary(), can improve such a dictionary's effectiveness at low cost. This PR changes how dictionary is created by calling the ZSTD ZDICT_finalizeDictionary() API instead of creating raw content dictionary (when max_dict_buffer_bytes > 0), and pass in all buffered uncompressed data blocks as samples. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9857 Test Plan: #### db_bench test for cpu/memory of compression+decompression and space saving on synthetic data: Set up: change the parameter [here](https://github.com/facebook/rocksdb/blob/fb9a167a55e0970b1ef6f67c1600c8d9c4c6114f/tools/db_bench_tool.cc#L1766) to 16384 to make synthetic data more compressible. ``` # linked local ZSTD with version 1.5.2 # DEBUG_LEVEL=0 ROCKSDB_NO_FBCODE=1 ROCKSDB_DISABLE_ZSTD=1 EXTRA_CXXFLAGS="-DZSTD_STATIC_LINKING_ONLY -DZSTD -I/data/users/changyubi/install/include/" EXTRA_LDFLAGS="-L/data/users/changyubi/install/lib/ -l:libzstd.a" make -j32 db_bench dict_bytes=16384 train_bytes=1048576 echo "========== No Dictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=0 -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=0 -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== Raw Content Dictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench_main -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench_main -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== FinalizeDictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== TrainDictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total # Result: TrainDictionary is much better on space saving, but FinalizeDictionary seems to use less memory. # before compression data size: 1.2GB dict_bytes=16384 max_dict_buffer_bytes = 1048576 space cpu/memory No Dictionary 468M 14.93user 1.00system 0:15.92elapsed 100%CPU (0avgtext+0avgdata 23904maxresident)k Raw Dictionary 251M 15.81user 0.80system 0:16.56elapsed 100%CPU (0avgtext+0avgdata 156808maxresident)k FinalizeDictionary 236M 11.93user 0.64system 0:12.56elapsed 100%CPU (0avgtext+0avgdata 89548maxresident)k TrainDictionary 84M 7.29user 0.45system 0:07.75elapsed 100%CPU (0avgtext+0avgdata 97288maxresident)k ``` #### Benchmark on 10 sample SST files for spacing saving and CPU time on compression: FinalizeDictionary is comparable to TrainDictionary in terms of space saving, and takes less time in compression. ``` dict_bytes=16384 train_bytes=1048576 for sst_file in `ls ../temp/myrock-sst/` do echo "********** $sst_file **********" echo "========== No Dictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD echo "========== Raw Content Dictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes echo "========== FinalizeDictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes --compression_zstd_max_train_bytes=$train_bytes --compression_use_zstd_finalize_dict echo "========== TrainDictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes --compression_zstd_max_train_bytes=$train_bytes done 010240.sst (Size/Time) 011029.sst 013184.sst 021552.sst 185054.sst 185137.sst 191666.sst 7560381.sst 7604174.sst 7635312.sst No Dictionary 28165569 / 2614419 32899411 / 2976832 32977848 / 3055542 31966329 / 2004590 33614351 / 1755877 33429029 / 1717042 33611933 / 1776936 33634045 / 2771417 33789721 / 2205414 33592194 / 388254 Raw Content Dictionary 28019950 / 2697961 33748665 / 3572422 33896373 / 3534701 26418431 / 2259658 28560825 / 1839168 28455030 / 1846039 28494319 / 1861349 32391599 / 3095649 33772142 / 2407843 33592230 / 474523 FinalizeDictionary 27896012 / 2650029 33763886 / 3719427 33904283 / 3552793 26008225 / 2198033 28111872 / 1869530 28014374 / 1789771 28047706 / 1848300 32296254 / 3204027 33698698 / 2381468 33592344 / 517433 TrainDictionary 28046089 / 2740037 33706480 / 3679019 33885741 / 3629351 25087123 / 2204558 27194353 / 1970207 27234229 / 1896811 27166710 / 1903119 32011041 / 3322315 32730692 / 2406146 33608631 / 570593 ``` #### Decompression/Read test: With FinalizeDictionary/TrainDictionary, some data structure used for decompression are in stored in dictionary, so they are expected to be faster in terms of decompression/reads. ``` dict_bytes=16384 train_bytes=1048576 echo "No Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=0 > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=0 2>&1 | grep MB/s echo "Raw Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes 2>&1 | grep MB/s echo "FinalizeDict" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false 2>&1 | grep MB/s echo "Train Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes 2>&1 | grep MB/s No Dictionary readrandom : 12.183 micros/op 82082 ops/sec 12.183 seconds 1000000 operations; 9.1 MB/s (1000000 of 1000000 found) Raw Dictionary readrandom : 12.314 micros/op 81205 ops/sec 12.314 seconds 1000000 operations; 9.0 MB/s (1000000 of 1000000 found) FinalizeDict readrandom : 9.787 micros/op 102180 ops/sec 9.787 seconds 1000000 operations; 11.3 MB/s (1000000 of 1000000 found) Train Dictionary readrandom : 9.698 micros/op 103108 ops/sec 9.699 seconds 1000000 operations; 11.4 MB/s (1000000 of 1000000 found) ``` Reviewed By: ajkr Differential Revision: D35720026 Pulled By: cbi42 fbshipit-source-id: 24d230fdff0fd28a1bb650658798f00dfcfb2a1f
2 years ago
} else if (i == kWithZSTDfinalizeDict) {
bytes_with_zstd_finalize_dict = total_sst_bytes;
} else if (i == kWithZSTDTrainedDict) {
bytes_with_zstd_trained_dict = total_sst_bytes;
}
for (size_t j = 0; j < kNumL0Files * (kL0FileBytes / kBlockSizeBytes);
j++) {
Reduce scope of compression dictionary to single SST (#4952) Summary: Our previous approach was to train one compression dictionary per compaction, using the first output SST to train a dictionary, and then applying it on subsequent SSTs in the same compaction. While this was great for minimizing CPU/memory/I/O overhead, it did not achieve good compression ratios in practice. In our most promising potential use case, moderate reductions in a dictionary's scope make a major difference on compression ratio. So, this PR changes compression dictionary to be scoped per-SST. It accepts the tradeoff during table building to use more memory and CPU. Important changes include: - The `BlockBasedTableBuilder` has a new state when dictionary compression is in-use: `kBuffered`. In that state it accumulates uncompressed data in-memory whenever `Add` is called. - After accumulating target file size bytes or calling `BlockBasedTableBuilder::Finish`, a `BlockBasedTableBuilder` moves to the `kUnbuffered` state. The transition (`EnterUnbuffered()`) involves sampling the buffered data, training a dictionary, and compressing/writing out all buffered data. In the `kUnbuffered` state, a `BlockBasedTableBuilder` behaves the same as before -- blocks are compressed/written out as soon as they fill up. - Samples are now whole uncompressed data blocks, except the final sample may be a partial data block so we don't breach the user's configured `max_dict_bytes` or `zstd_max_train_bytes`. The dictionary trainer is supposed to work better when we pass it real units of compression. Previously we were passing 64-byte KV samples which was not realistic. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4952 Differential Revision: D13967980 Pulled By: ajkr fbshipit-source-id: 82bea6f7537e1529c7a1a4cdee84585f5949300f
6 years ago
ASSERT_EQ(seq_datas[(j / 10) % 10], Get(1, Key(static_cast<int>(j))));
}
if (i == kWithDict) {
ASSERT_GT(bytes_without_dict, bytes_with_dict);
Support using ZDICT_finalizeDictionary to generate zstd dictionary (#9857) Summary: An untrained dictionary is currently simply the concatenation of several samples. The ZSTD API, ZDICT_finalizeDictionary(), can improve such a dictionary's effectiveness at low cost. This PR changes how dictionary is created by calling the ZSTD ZDICT_finalizeDictionary() API instead of creating raw content dictionary (when max_dict_buffer_bytes > 0), and pass in all buffered uncompressed data blocks as samples. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9857 Test Plan: #### db_bench test for cpu/memory of compression+decompression and space saving on synthetic data: Set up: change the parameter [here](https://github.com/facebook/rocksdb/blob/fb9a167a55e0970b1ef6f67c1600c8d9c4c6114f/tools/db_bench_tool.cc#L1766) to 16384 to make synthetic data more compressible. ``` # linked local ZSTD with version 1.5.2 # DEBUG_LEVEL=0 ROCKSDB_NO_FBCODE=1 ROCKSDB_DISABLE_ZSTD=1 EXTRA_CXXFLAGS="-DZSTD_STATIC_LINKING_ONLY -DZSTD -I/data/users/changyubi/install/include/" EXTRA_LDFLAGS="-L/data/users/changyubi/install/lib/ -l:libzstd.a" make -j32 db_bench dict_bytes=16384 train_bytes=1048576 echo "========== No Dictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=0 -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=0 -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== Raw Content Dictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench_main -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench_main -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== FinalizeDictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total echo "========== TrainDictionary ==========" TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=filluniquerandom,compact -num=10000000 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -block_size=4096 -max_background_jobs=24 -memtablerep=vector -allow_concurrent_memtable_write=false -disable_wal=true -max_write_buffer_number=8 >/dev/null 2>&1 TEST_TMPDIR=/dev/shm /usr/bin/time ./db_bench -use_existing_db=true -benchmarks=compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -block_size=4096 2>&1 | grep elapsed du -hc /dev/shm/dbbench/*sst | grep total # Result: TrainDictionary is much better on space saving, but FinalizeDictionary seems to use less memory. # before compression data size: 1.2GB dict_bytes=16384 max_dict_buffer_bytes = 1048576 space cpu/memory No Dictionary 468M 14.93user 1.00system 0:15.92elapsed 100%CPU (0avgtext+0avgdata 23904maxresident)k Raw Dictionary 251M 15.81user 0.80system 0:16.56elapsed 100%CPU (0avgtext+0avgdata 156808maxresident)k FinalizeDictionary 236M 11.93user 0.64system 0:12.56elapsed 100%CPU (0avgtext+0avgdata 89548maxresident)k TrainDictionary 84M 7.29user 0.45system 0:07.75elapsed 100%CPU (0avgtext+0avgdata 97288maxresident)k ``` #### Benchmark on 10 sample SST files for spacing saving and CPU time on compression: FinalizeDictionary is comparable to TrainDictionary in terms of space saving, and takes less time in compression. ``` dict_bytes=16384 train_bytes=1048576 for sst_file in `ls ../temp/myrock-sst/` do echo "********** $sst_file **********" echo "========== No Dictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD echo "========== Raw Content Dictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes echo "========== FinalizeDictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes --compression_zstd_max_train_bytes=$train_bytes --compression_use_zstd_finalize_dict echo "========== TrainDictionary ==========" ./sst_dump --file="../temp/myrock-sst/$sst_file" --command=recompress --compression_level_from=6 --compression_level_to=6 --compression_types=kZSTD --compression_max_dict_bytes=$dict_bytes --compression_zstd_max_train_bytes=$train_bytes done 010240.sst (Size/Time) 011029.sst 013184.sst 021552.sst 185054.sst 185137.sst 191666.sst 7560381.sst 7604174.sst 7635312.sst No Dictionary 28165569 / 2614419 32899411 / 2976832 32977848 / 3055542 31966329 / 2004590 33614351 / 1755877 33429029 / 1717042 33611933 / 1776936 33634045 / 2771417 33789721 / 2205414 33592194 / 388254 Raw Content Dictionary 28019950 / 2697961 33748665 / 3572422 33896373 / 3534701 26418431 / 2259658 28560825 / 1839168 28455030 / 1846039 28494319 / 1861349 32391599 / 3095649 33772142 / 2407843 33592230 / 474523 FinalizeDictionary 27896012 / 2650029 33763886 / 3719427 33904283 / 3552793 26008225 / 2198033 28111872 / 1869530 28014374 / 1789771 28047706 / 1848300 32296254 / 3204027 33698698 / 2381468 33592344 / 517433 TrainDictionary 28046089 / 2740037 33706480 / 3679019 33885741 / 3629351 25087123 / 2204558 27194353 / 1970207 27234229 / 1896811 27166710 / 1903119 32011041 / 3322315 32730692 / 2406146 33608631 / 570593 ``` #### Decompression/Read test: With FinalizeDictionary/TrainDictionary, some data structure used for decompression are in stored in dictionary, so they are expected to be faster in terms of decompression/reads. ``` dict_bytes=16384 train_bytes=1048576 echo "No Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=0 > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=0 2>&1 | grep MB/s echo "Raw Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes 2>&1 | grep MB/s echo "FinalizeDict" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes -compression_use_zstd_dict_trainer=false 2>&1 | grep MB/s echo "Train Dictionary" TEST_TMPDIR=/dev/shm/ ./db_bench -benchmarks=filluniquerandom,compact -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes > /dev/null 2>&1 TEST_TMPDIR=/dev/shm/ ./db_bench -use_existing_db=true -benchmarks=readrandom -cache_size=0 -compression_type=zstd -compression_max_dict_bytes=$dict_bytes -compression_zstd_max_train_bytes=$train_bytes 2>&1 | grep MB/s No Dictionary readrandom : 12.183 micros/op 82082 ops/sec 12.183 seconds 1000000 operations; 9.1 MB/s (1000000 of 1000000 found) Raw Dictionary readrandom : 12.314 micros/op 81205 ops/sec 12.314 seconds 1000000 operations; 9.0 MB/s (1000000 of 1000000 found) FinalizeDict readrandom : 9.787 micros/op 102180 ops/sec 9.787 seconds 1000000 operations; 11.3 MB/s (1000000 of 1000000 found) Train Dictionary readrandom : 9.698 micros/op 103108 ops/sec 9.699 seconds 1000000 operations; 11.4 MB/s (1000000 of 1000000 found) ``` Reviewed By: ajkr Differential Revision: D35720026 Pulled By: cbi42 fbshipit-source-id: 24d230fdff0fd28a1bb650658798f00dfcfb2a1f
2 years ago
} else if (i == kWithZSTDTrainedDict) {
// In zstd compression, it is sometimes possible that using a finalized
// dictionary does not get as good a compression ratio as raw content
// dictionary. But using a dictionary should always get better
// compression ratio than not using one.
ASSERT_TRUE(bytes_with_dict > bytes_with_zstd_finalize_dict ||
bytes_without_dict > bytes_with_zstd_finalize_dict);
} else if (i == kWithZSTDTrainedDict) {
// In zstd compression, it is sometimes possible that using a trained
// dictionary does not get as good a compression ratio as without
// training.
// But using a dictionary (with or without training) should always get
// better compression ratio than not using one.
ASSERT_TRUE(bytes_with_dict > bytes_with_zstd_trained_dict ||
bytes_without_dict > bytes_with_zstd_trained_dict);
}
DestroyAndReopen(options);
}
}
}
Reduce scope of compression dictionary to single SST (#4952) Summary: Our previous approach was to train one compression dictionary per compaction, using the first output SST to train a dictionary, and then applying it on subsequent SSTs in the same compaction. While this was great for minimizing CPU/memory/I/O overhead, it did not achieve good compression ratios in practice. In our most promising potential use case, moderate reductions in a dictionary's scope make a major difference on compression ratio. So, this PR changes compression dictionary to be scoped per-SST. It accepts the tradeoff during table building to use more memory and CPU. Important changes include: - The `BlockBasedTableBuilder` has a new state when dictionary compression is in-use: `kBuffered`. In that state it accumulates uncompressed data in-memory whenever `Add` is called. - After accumulating target file size bytes or calling `BlockBasedTableBuilder::Finish`, a `BlockBasedTableBuilder` moves to the `kUnbuffered` state. The transition (`EnterUnbuffered()`) involves sampling the buffered data, training a dictionary, and compressing/writing out all buffered data. In the `kUnbuffered` state, a `BlockBasedTableBuilder` behaves the same as before -- blocks are compressed/written out as soon as they fill up. - Samples are now whole uncompressed data blocks, except the final sample may be a partial data block so we don't breach the user's configured `max_dict_bytes` or `zstd_max_train_bytes`. The dictionary trainer is supposed to work better when we pass it real units of compression. Previously we were passing 64-byte KV samples which was not realistic. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4952 Differential Revision: D13967980 Pulled By: ajkr fbshipit-source-id: 82bea6f7537e1529c7a1a4cdee84585f5949300f
6 years ago
TEST_F(DBTest2, PresetCompressionDictLocality) {
if (!ZSTD_Supported()) {
return;
}
// Verifies that compression dictionary is generated from local data. The
// verification simply checks all output SSTs have different compression
// dictionaries. We do not verify effectiveness as that'd likely be flaky in
// the future.
const int kNumEntriesPerFile = 1 << 10; // 1KB
const int kNumBytesPerEntry = 1 << 10; // 1KB
const int kNumFiles = 4;
Options options = CurrentOptions();
options.compression = kZSTD;
options.compression_opts.max_dict_bytes = 1 << 14; // 16KB
options.compression_opts.zstd_max_train_bytes = 1 << 18; // 256KB
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
Reduce scope of compression dictionary to single SST (#4952) Summary: Our previous approach was to train one compression dictionary per compaction, using the first output SST to train a dictionary, and then applying it on subsequent SSTs in the same compaction. While this was great for minimizing CPU/memory/I/O overhead, it did not achieve good compression ratios in practice. In our most promising potential use case, moderate reductions in a dictionary's scope make a major difference on compression ratio. So, this PR changes compression dictionary to be scoped per-SST. It accepts the tradeoff during table building to use more memory and CPU. Important changes include: - The `BlockBasedTableBuilder` has a new state when dictionary compression is in-use: `kBuffered`. In that state it accumulates uncompressed data in-memory whenever `Add` is called. - After accumulating target file size bytes or calling `BlockBasedTableBuilder::Finish`, a `BlockBasedTableBuilder` moves to the `kUnbuffered` state. The transition (`EnterUnbuffered()`) involves sampling the buffered data, training a dictionary, and compressing/writing out all buffered data. In the `kUnbuffered` state, a `BlockBasedTableBuilder` behaves the same as before -- blocks are compressed/written out as soon as they fill up. - Samples are now whole uncompressed data blocks, except the final sample may be a partial data block so we don't breach the user's configured `max_dict_bytes` or `zstd_max_train_bytes`. The dictionary trainer is supposed to work better when we pass it real units of compression. Previously we were passing 64-byte KV samples which was not realistic. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4952 Differential Revision: D13967980 Pulled By: ajkr fbshipit-source-id: 82bea6f7537e1529c7a1a4cdee84585f5949300f
6 years ago
options.target_file_size_base = kNumEntriesPerFile * kNumBytesPerEntry;
BlockBasedTableOptions table_options;
table_options.cache_index_and_filter_blocks = true;
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
Reduce scope of compression dictionary to single SST (#4952) Summary: Our previous approach was to train one compression dictionary per compaction, using the first output SST to train a dictionary, and then applying it on subsequent SSTs in the same compaction. While this was great for minimizing CPU/memory/I/O overhead, it did not achieve good compression ratios in practice. In our most promising potential use case, moderate reductions in a dictionary's scope make a major difference on compression ratio. So, this PR changes compression dictionary to be scoped per-SST. It accepts the tradeoff during table building to use more memory and CPU. Important changes include: - The `BlockBasedTableBuilder` has a new state when dictionary compression is in-use: `kBuffered`. In that state it accumulates uncompressed data in-memory whenever `Add` is called. - After accumulating target file size bytes or calling `BlockBasedTableBuilder::Finish`, a `BlockBasedTableBuilder` moves to the `kUnbuffered` state. The transition (`EnterUnbuffered()`) involves sampling the buffered data, training a dictionary, and compressing/writing out all buffered data. In the `kUnbuffered` state, a `BlockBasedTableBuilder` behaves the same as before -- blocks are compressed/written out as soon as they fill up. - Samples are now whole uncompressed data blocks, except the final sample may be a partial data block so we don't breach the user's configured `max_dict_bytes` or `zstd_max_train_bytes`. The dictionary trainer is supposed to work better when we pass it real units of compression. Previously we were passing 64-byte KV samples which was not realistic. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4952 Differential Revision: D13967980 Pulled By: ajkr fbshipit-source-id: 82bea6f7537e1529c7a1a4cdee84585f5949300f
6 years ago
Reopen(options);
Random rnd(301);
for (int i = 0; i < kNumFiles; ++i) {
for (int j = 0; j < kNumEntriesPerFile; ++j) {
ASSERT_OK(Put(Key(i * kNumEntriesPerFile + j),
rnd.RandomString(kNumBytesPerEntry)));
Reduce scope of compression dictionary to single SST (#4952) Summary: Our previous approach was to train one compression dictionary per compaction, using the first output SST to train a dictionary, and then applying it on subsequent SSTs in the same compaction. While this was great for minimizing CPU/memory/I/O overhead, it did not achieve good compression ratios in practice. In our most promising potential use case, moderate reductions in a dictionary's scope make a major difference on compression ratio. So, this PR changes compression dictionary to be scoped per-SST. It accepts the tradeoff during table building to use more memory and CPU. Important changes include: - The `BlockBasedTableBuilder` has a new state when dictionary compression is in-use: `kBuffered`. In that state it accumulates uncompressed data in-memory whenever `Add` is called. - After accumulating target file size bytes or calling `BlockBasedTableBuilder::Finish`, a `BlockBasedTableBuilder` moves to the `kUnbuffered` state. The transition (`EnterUnbuffered()`) involves sampling the buffered data, training a dictionary, and compressing/writing out all buffered data. In the `kUnbuffered` state, a `BlockBasedTableBuilder` behaves the same as before -- blocks are compressed/written out as soon as they fill up. - Samples are now whole uncompressed data blocks, except the final sample may be a partial data block so we don't breach the user's configured `max_dict_bytes` or `zstd_max_train_bytes`. The dictionary trainer is supposed to work better when we pass it real units of compression. Previously we were passing 64-byte KV samples which was not realistic. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4952 Differential Revision: D13967980 Pulled By: ajkr fbshipit-source-id: 82bea6f7537e1529c7a1a4cdee84585f5949300f
6 years ago
}
ASSERT_OK(Flush());
MoveFilesToLevel(1);
ASSERT_EQ(NumTableFilesAtLevel(1), i + 1);
}
// Store all the dictionaries generated during a full compaction.
std::vector<std::string> compression_dicts;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
Reduce scope of compression dictionary to single SST (#4952) Summary: Our previous approach was to train one compression dictionary per compaction, using the first output SST to train a dictionary, and then applying it on subsequent SSTs in the same compaction. While this was great for minimizing CPU/memory/I/O overhead, it did not achieve good compression ratios in practice. In our most promising potential use case, moderate reductions in a dictionary's scope make a major difference on compression ratio. So, this PR changes compression dictionary to be scoped per-SST. It accepts the tradeoff during table building to use more memory and CPU. Important changes include: - The `BlockBasedTableBuilder` has a new state when dictionary compression is in-use: `kBuffered`. In that state it accumulates uncompressed data in-memory whenever `Add` is called. - After accumulating target file size bytes or calling `BlockBasedTableBuilder::Finish`, a `BlockBasedTableBuilder` moves to the `kUnbuffered` state. The transition (`EnterUnbuffered()`) involves sampling the buffered data, training a dictionary, and compressing/writing out all buffered data. In the `kUnbuffered` state, a `BlockBasedTableBuilder` behaves the same as before -- blocks are compressed/written out as soon as they fill up. - Samples are now whole uncompressed data blocks, except the final sample may be a partial data block so we don't breach the user's configured `max_dict_bytes` or `zstd_max_train_bytes`. The dictionary trainer is supposed to work better when we pass it real units of compression. Previously we were passing 64-byte KV samples which was not realistic. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4952 Differential Revision: D13967980 Pulled By: ajkr fbshipit-source-id: 82bea6f7537e1529c7a1a4cdee84585f5949300f
6 years ago
"BlockBasedTableBuilder::WriteCompressionDictBlock:RawDict",
[&](void* arg) {
compression_dicts.emplace_back(static_cast<Slice*>(arg)->ToString());
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
Reduce scope of compression dictionary to single SST (#4952) Summary: Our previous approach was to train one compression dictionary per compaction, using the first output SST to train a dictionary, and then applying it on subsequent SSTs in the same compaction. While this was great for minimizing CPU/memory/I/O overhead, it did not achieve good compression ratios in practice. In our most promising potential use case, moderate reductions in a dictionary's scope make a major difference on compression ratio. So, this PR changes compression dictionary to be scoped per-SST. It accepts the tradeoff during table building to use more memory and CPU. Important changes include: - The `BlockBasedTableBuilder` has a new state when dictionary compression is in-use: `kBuffered`. In that state it accumulates uncompressed data in-memory whenever `Add` is called. - After accumulating target file size bytes or calling `BlockBasedTableBuilder::Finish`, a `BlockBasedTableBuilder` moves to the `kUnbuffered` state. The transition (`EnterUnbuffered()`) involves sampling the buffered data, training a dictionary, and compressing/writing out all buffered data. In the `kUnbuffered` state, a `BlockBasedTableBuilder` behaves the same as before -- blocks are compressed/written out as soon as they fill up. - Samples are now whole uncompressed data blocks, except the final sample may be a partial data block so we don't breach the user's configured `max_dict_bytes` or `zstd_max_train_bytes`. The dictionary trainer is supposed to work better when we pass it real units of compression. Previously we were passing 64-byte KV samples which was not realistic. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4952 Differential Revision: D13967980 Pulled By: ajkr fbshipit-source-id: 82bea6f7537e1529c7a1a4cdee84585f5949300f
6 years ago
CompactRangeOptions compact_range_opts;
compact_range_opts.bottommost_level_compaction =
BottommostLevelCompaction::kForceOptimized;
Reduce scope of compression dictionary to single SST (#4952) Summary: Our previous approach was to train one compression dictionary per compaction, using the first output SST to train a dictionary, and then applying it on subsequent SSTs in the same compaction. While this was great for minimizing CPU/memory/I/O overhead, it did not achieve good compression ratios in practice. In our most promising potential use case, moderate reductions in a dictionary's scope make a major difference on compression ratio. So, this PR changes compression dictionary to be scoped per-SST. It accepts the tradeoff during table building to use more memory and CPU. Important changes include: - The `BlockBasedTableBuilder` has a new state when dictionary compression is in-use: `kBuffered`. In that state it accumulates uncompressed data in-memory whenever `Add` is called. - After accumulating target file size bytes or calling `BlockBasedTableBuilder::Finish`, a `BlockBasedTableBuilder` moves to the `kUnbuffered` state. The transition (`EnterUnbuffered()`) involves sampling the buffered data, training a dictionary, and compressing/writing out all buffered data. In the `kUnbuffered` state, a `BlockBasedTableBuilder` behaves the same as before -- blocks are compressed/written out as soon as they fill up. - Samples are now whole uncompressed data blocks, except the final sample may be a partial data block so we don't breach the user's configured `max_dict_bytes` or `zstd_max_train_bytes`. The dictionary trainer is supposed to work better when we pass it real units of compression. Previously we were passing 64-byte KV samples which was not realistic. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4952 Differential Revision: D13967980 Pulled By: ajkr fbshipit-source-id: 82bea6f7537e1529c7a1a4cdee84585f5949300f
6 years ago
ASSERT_OK(db_->CompactRange(compact_range_opts, nullptr, nullptr));
// Dictionary compression should not be so good as to compress four totally
// random files into one. If it does then there's probably something wrong
// with the test.
ASSERT_GT(NumTableFilesAtLevel(1), 1);
// Furthermore, there should be one compression dictionary generated per file.
// And they should all be different from each other.
ASSERT_EQ(NumTableFilesAtLevel(1),
static_cast<int>(compression_dicts.size()));
for (size_t i = 1; i < compression_dicts.size(); ++i) {
std::string& a = compression_dicts[i - 1];
std::string& b = compression_dicts[i];
size_t alen = a.size();
size_t blen = b.size();
ASSERT_TRUE(alen != blen || memcmp(a.data(), b.data(), alen) != 0);
}
}
class PresetCompressionDictTest
: public DBTestBase,
public testing::WithParamInterface<std::tuple<CompressionType, bool>> {
public:
PresetCompressionDictTest()
: DBTestBase("db_test2", false /* env_do_fsync */),
compression_type_(std::get<0>(GetParam())),
bottommost_(std::get<1>(GetParam())) {}
protected:
const CompressionType compression_type_;
const bool bottommost_;
};
INSTANTIATE_TEST_CASE_P(
DBTest2, PresetCompressionDictTest,
::testing::Combine(::testing::ValuesIn(GetSupportedDictCompressions()),
::testing::Bool()));
TEST_P(PresetCompressionDictTest, Flush) {
// Verifies that dictionary is generated and written during flush only when
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
// `ColumnFamilyOptions::compression` enables dictionary. Also verifies the
// size of the dictionary is within expectations according to the limit on
// buffering set by `CompressionOptions::max_dict_buffer_bytes`.
const size_t kValueLen = 256;
const size_t kKeysPerFile = 1 << 10;
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
const size_t kDictLen = 16 << 10;
const size_t kBlockLen = 4 << 10;
Options options = CurrentOptions();
if (bottommost_) {
options.bottommost_compression = compression_type_;
options.bottommost_compression_opts.enabled = true;
options.bottommost_compression_opts.max_dict_bytes = kDictLen;
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
options.bottommost_compression_opts.max_dict_buffer_bytes = kBlockLen;
} else {
options.compression = compression_type_;
options.compression_opts.max_dict_bytes = kDictLen;
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
options.compression_opts.max_dict_buffer_bytes = kBlockLen;
}
options.memtable_factory.reset(test::NewSpecialSkipListFactory(kKeysPerFile));
options.statistics = CreateDBStatistics();
BlockBasedTableOptions bbto;
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
bbto.block_size = kBlockLen;
bbto.cache_index_and_filter_blocks = true;
options.table_factory.reset(NewBlockBasedTableFactory(bbto));
Reopen(options);
Random rnd(301);
for (size_t i = 0; i <= kKeysPerFile; ++i) {
ASSERT_OK(Put(Key(static_cast<int>(i)), rnd.RandomString(kValueLen)));
}
ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable());
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
// We can use `BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT` to detect whether a
// compression dictionary exists since dictionaries would be preloaded when
// the flush finishes.
if (bottommost_) {
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
// Flush is never considered bottommost. This should change in the future
// since flushed files may have nothing underneath them, like the one in
// this test case.
ASSERT_EQ(
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT),
0);
} else {
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
ASSERT_GT(
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT),
0);
// TODO(ajkr): fix the below assertion to work with ZSTD. The expectation on
// number of bytes needs to be adjusted in case the cached block is in
// ZSTD's digested dictionary format.
if (compression_type_ != kZSTD &&
compression_type_ != kZSTDNotFinalCompression) {
// Although we limited buffering to `kBlockLen`, there may be up to two
// blocks of data included in the dictionary since we only check limit
// after each block is built.
ASSERT_LE(TestGetTickerCount(options,
BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT),
2 * kBlockLen);
}
}
}
TEST_P(PresetCompressionDictTest, CompactNonBottommost) {
// Verifies that dictionary is generated and written during compaction to
// non-bottommost level only when `ColumnFamilyOptions::compression` enables
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
// dictionary. Also verifies the size of the dictionary is within expectations
// according to the limit on buffering set by
// `CompressionOptions::max_dict_buffer_bytes`.
const size_t kValueLen = 256;
const size_t kKeysPerFile = 1 << 10;
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
const size_t kDictLen = 16 << 10;
const size_t kBlockLen = 4 << 10;
Options options = CurrentOptions();
if (bottommost_) {
options.bottommost_compression = compression_type_;
options.bottommost_compression_opts.enabled = true;
options.bottommost_compression_opts.max_dict_bytes = kDictLen;
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
options.bottommost_compression_opts.max_dict_buffer_bytes = kBlockLen;
} else {
options.compression = compression_type_;
options.compression_opts.max_dict_bytes = kDictLen;
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
options.compression_opts.max_dict_buffer_bytes = kBlockLen;
}
options.disable_auto_compactions = true;
options.statistics = CreateDBStatistics();
BlockBasedTableOptions bbto;
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
bbto.block_size = kBlockLen;
bbto.cache_index_and_filter_blocks = true;
options.table_factory.reset(NewBlockBasedTableFactory(bbto));
Reopen(options);
Random rnd(301);
for (size_t j = 0; j <= kKeysPerFile; ++j) {
ASSERT_OK(Put(Key(static_cast<int>(j)), rnd.RandomString(kValueLen)));
}
ASSERT_OK(Flush());
MoveFilesToLevel(2);
for (int i = 0; i < 2; ++i) {
for (size_t j = 0; j <= kKeysPerFile; ++j) {
ASSERT_OK(Put(Key(static_cast<int>(j)), rnd.RandomString(kValueLen)));
}
ASSERT_OK(Flush());
}
#ifndef ROCKSDB_LITE
ASSERT_EQ("2,0,1", FilesPerLevel(0));
#endif // ROCKSDB_LITE
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
uint64_t prev_compression_dict_bytes_inserted =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT);
// This L0->L1 compaction merges the two L0 files into L1. The produced L1
// file is not bottommost due to the existing L2 file covering the same key-
// range.
ASSERT_OK(dbfull()->TEST_CompactRange(0, nullptr, nullptr));
#ifndef ROCKSDB_LITE
ASSERT_EQ("0,1,1", FilesPerLevel(0));
#endif // ROCKSDB_LITE
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
// We can use `BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT` to detect whether a
// compression dictionary exists since dictionaries would be preloaded when
// the compaction finishes.
if (bottommost_) {
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
ASSERT_EQ(
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT),
prev_compression_dict_bytes_inserted);
} else {
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
ASSERT_GT(
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT),
prev_compression_dict_bytes_inserted);
// TODO(ajkr): fix the below assertion to work with ZSTD. The expectation on
// number of bytes needs to be adjusted in case the cached block is in
// ZSTD's digested dictionary format.
if (compression_type_ != kZSTD &&
compression_type_ != kZSTDNotFinalCompression) {
// Although we limited buffering to `kBlockLen`, there may be up to two
// blocks of data included in the dictionary since we only check limit
// after each block is built.
ASSERT_LE(TestGetTickerCount(options,
BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT),
prev_compression_dict_bytes_inserted + 2 * kBlockLen);
}
}
}
TEST_P(PresetCompressionDictTest, CompactBottommost) {
// Verifies that dictionary is generated and written during compaction to
// non-bottommost level only when either `ColumnFamilyOptions::compression` or
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
// `ColumnFamilyOptions::bottommost_compression` enables dictionary. Also
// verifies the size of the dictionary is within expectations according to the
// limit on buffering set by `CompressionOptions::max_dict_buffer_bytes`.
const size_t kValueLen = 256;
const size_t kKeysPerFile = 1 << 10;
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
const size_t kDictLen = 16 << 10;
const size_t kBlockLen = 4 << 10;
Options options = CurrentOptions();
if (bottommost_) {
options.bottommost_compression = compression_type_;
options.bottommost_compression_opts.enabled = true;
options.bottommost_compression_opts.max_dict_bytes = kDictLen;
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
options.bottommost_compression_opts.max_dict_buffer_bytes = kBlockLen;
} else {
options.compression = compression_type_;
options.compression_opts.max_dict_bytes = kDictLen;
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
options.compression_opts.max_dict_buffer_bytes = kBlockLen;
}
options.disable_auto_compactions = true;
options.statistics = CreateDBStatistics();
BlockBasedTableOptions bbto;
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
bbto.block_size = kBlockLen;
bbto.cache_index_and_filter_blocks = true;
options.table_factory.reset(NewBlockBasedTableFactory(bbto));
Reopen(options);
Random rnd(301);
for (int i = 0; i < 2; ++i) {
for (size_t j = 0; j <= kKeysPerFile; ++j) {
ASSERT_OK(Put(Key(static_cast<int>(j)), rnd.RandomString(kValueLen)));
}
ASSERT_OK(Flush());
}
#ifndef ROCKSDB_LITE
ASSERT_EQ("2", FilesPerLevel(0));
#endif // ROCKSDB_LITE
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
uint64_t prev_compression_dict_bytes_inserted =
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT);
CompactRangeOptions cro;
ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
#ifndef ROCKSDB_LITE
ASSERT_EQ("0,1", FilesPerLevel(0));
#endif // ROCKSDB_LITE
Limit buffering for collecting samples for compression dictionary (#7970) Summary: For dictionary compression, we need to collect some representative samples of the data to be compressed, which we use to either generate or train (when `CompressionOptions::zstd_max_train_bytes > 0`) a dictionary. Previously, the strategy was to buffer all the data blocks during flush, and up to the target file size during compaction. That strategy allowed us to randomly pick samples from as wide a range as possible that'd be guaranteed to land in a single output file. However, some users try to make huge files in memory-constrained environments, where this strategy can cause OOM. This PR introduces an option, `CompressionOptions::max_dict_buffer_bytes`, that limits how much data blocks are buffered before we switch to unbuffered mode (which means creating the per-SST dictionary, writing out the buffered data, and compressing/writing new blocks as soon as they are built). It is not strict as we currently buffer more than just data blocks -- also keys are buffered. But it does make a step towards giving users predictable memory usage. Related changes include: - Changed sampling for dictionary compression to select unique data blocks when there is limited availability of data blocks - Made use of `BlockBuilder::SwapAndReset()` to save an allocation+memcpy when buffering data blocks for building a dictionary - Changed `ParseBoolean()` to accept an input containing characters after the boolean. This is necessary since, with this PR, a value for `CompressionOptions::enabled` is no longer necessarily the final component in the `CompressionOptions` string. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7970 Test Plan: - updated `CompressionOptions` unit tests to verify limit is respected (to the extent expected in the current implementation) in various scenarios of flush/compaction to bottommost/non-bottommost level - looked at jemalloc heap profiles right before and after switching to unbuffered mode during flush/compaction. Verified memory usage in buffering is proportional to the limit set. Reviewed By: pdillinger Differential Revision: D26467994 Pulled By: ajkr fbshipit-source-id: 3da4ef9fba59974e4ef40e40c01611002c861465
4 years ago
ASSERT_GT(
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT),
prev_compression_dict_bytes_inserted);
// TODO(ajkr): fix the below assertion to work with ZSTD. The expectation on
// number of bytes needs to be adjusted in case the cached block is in ZSTD's
// digested dictionary format.
if (compression_type_ != kZSTD &&
compression_type_ != kZSTDNotFinalCompression) {
// Although we limited buffering to `kBlockLen`, there may be up to two
// blocks of data included in the dictionary since we only check limit after
// each block is built.
ASSERT_LE(
TestGetTickerCount(options, BLOCK_CACHE_COMPRESSION_DICT_BYTES_INSERT),
prev_compression_dict_bytes_inserted + 2 * kBlockLen);
}
}
class CompactionCompressionListener : public EventListener {
public:
explicit CompactionCompressionListener(Options* db_options)
: db_options_(db_options) {}
void OnCompactionCompleted(DB* db, const CompactionJobInfo& ci) override {
// Figure out last level with files
int bottommost_level = 0;
for (int level = 0; level < db->NumberLevels(); level++) {
std::string files_at_level;
ASSERT_TRUE(
db->GetProperty("rocksdb.num-files-at-level" + std::to_string(level),
&files_at_level));
if (files_at_level != "0") {
bottommost_level = level;
}
}
if (db_options_->bottommost_compression != kDisableCompressionOption &&
ci.output_level == bottommost_level) {
ASSERT_EQ(ci.compression, db_options_->bottommost_compression);
} else if (db_options_->compression_per_level.size() != 0) {
ASSERT_EQ(ci.compression,
db_options_->compression_per_level[ci.output_level]);
} else {
ASSERT_EQ(ci.compression, db_options_->compression);
}
max_level_checked = std::max(max_level_checked, ci.output_level);
}
int max_level_checked = 0;
const Options* db_options_;
};
enum CompressionFailureType {
kTestCompressionFail,
kTestDecompressionFail,
kTestDecompressionCorruption
};
class CompressionFailuresTest
: public DBTest2,
public testing::WithParamInterface<std::tuple<
CompressionFailureType, CompressionType, uint32_t, uint32_t>> {
public:
CompressionFailuresTest() {
std::tie(compression_failure_type_, compression_type_,
compression_max_dict_bytes_, compression_parallel_threads_) =
GetParam();
}
CompressionFailureType compression_failure_type_ = kTestCompressionFail;
CompressionType compression_type_ = kNoCompression;
uint32_t compression_max_dict_bytes_ = 0;
uint32_t compression_parallel_threads_ = 0;
};
INSTANTIATE_TEST_CASE_P(
DBTest2, CompressionFailuresTest,
::testing::Combine(::testing::Values(kTestCompressionFail,
kTestDecompressionFail,
kTestDecompressionCorruption),
::testing::ValuesIn(GetSupportedCompressions()),
::testing::Values(0, 10), ::testing::Values(1, 4)));
TEST_P(CompressionFailuresTest, CompressionFailures) {
if (compression_type_ == kNoCompression) {
return;
}
Options options = CurrentOptions();
options.level0_file_num_compaction_trigger = 2;
options.max_bytes_for_level_base = 1024;
options.max_bytes_for_level_multiplier = 2;
options.num_levels = 7;
options.max_background_compactions = 1;
options.target_file_size_base = 512;
BlockBasedTableOptions table_options;
table_options.block_size = 512;
table_options.verify_compression = true;
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.compression = compression_type_;
options.compression_opts.parallel_threads = compression_parallel_threads_;
options.compression_opts.max_dict_bytes = compression_max_dict_bytes_;
options.bottommost_compression_opts.parallel_threads =
compression_parallel_threads_;
options.bottommost_compression_opts.max_dict_bytes =
compression_max_dict_bytes_;
if (compression_failure_type_ == kTestCompressionFail) {
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"CompressData:TamperWithReturnValue", [](void* arg) {
bool* ret = static_cast<bool*>(arg);
*ret = false;
});
} else if (compression_failure_type_ == kTestDecompressionFail) {
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
Refactor to avoid confusing "raw block" (#10408) Summary: We have a lot of confusing code because of mixed, sometimes completely opposite uses of of the term "raw block" or "raw contents", sometimes within the same source file. For example, in `BlockBasedTableBuilder`, `raw_block_contents` and `raw_size` generally referred to uncompressed block contents and size, while `WriteRawBlock` referred to writing a block that is already compressed if it is going to be. Meanwhile, in `BlockBasedTable`, `raw_block_contents` either referred to a (maybe compressed) block with trailer, or a maybe compressed block maybe without trailer. (Note: left as follow-up work to use C++ typing to better sort out the various kinds of BlockContents.) This change primarily tries to apply some consistent terminology around the kinds of block representations, avoiding the unclear "raw". (Any meaning of "raw" assumes some bias toward the storage layer or toward the logical data layer.) Preferred terminology: * **Serialized block** - bytes that go into storage. For block-based table (usually the case) this includes the block trailer. WART: block `size` may or may not include the trailer; need to be clear about whether it does or not. * **Maybe compressed block** - like a serialized block, but without the trailer (or no promise of including a trailer). Must be accompanied by a CompressionType. * **Uncompressed block** - "payload" bytes that are either stored with no compression, used as input to compression function, or result of decompression function. * **Parsed block** - an in-memory form of a block in block cache, as it is used by the table reader. Different C++ types are used depending on the block type (see block_like_traits.h). Other refactorings: * Misc corrections/improvements of internal API comments * Remove a few misleading / unhelpful / redundant comments. * Use move semantics in some places to simplify contracts * Use better parameter names to indicate which parameters are used for outputs * Remove some extraneous `extern` * Various clean-ups to `CacheDumperImpl` (mostly unnecessary code) Pull Request resolved: https://github.com/facebook/rocksdb/pull/10408 Test Plan: existing tests Reviewed By: akankshamahajan15 Differential Revision: D38172617 Pulled By: pdillinger fbshipit-source-id: ccb99299f324ac5ca46996d34c5089621a4f260c
2 years ago
"UncompressBlockData:TamperWithReturnValue", [](void* arg) {
Status* ret = static_cast<Status*>(arg);
ASSERT_OK(*ret);
*ret = Status::Corruption("kTestDecompressionFail");
});
} else if (compression_failure_type_ == kTestDecompressionCorruption) {
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
Refactor to avoid confusing "raw block" (#10408) Summary: We have a lot of confusing code because of mixed, sometimes completely opposite uses of of the term "raw block" or "raw contents", sometimes within the same source file. For example, in `BlockBasedTableBuilder`, `raw_block_contents` and `raw_size` generally referred to uncompressed block contents and size, while `WriteRawBlock` referred to writing a block that is already compressed if it is going to be. Meanwhile, in `BlockBasedTable`, `raw_block_contents` either referred to a (maybe compressed) block with trailer, or a maybe compressed block maybe without trailer. (Note: left as follow-up work to use C++ typing to better sort out the various kinds of BlockContents.) This change primarily tries to apply some consistent terminology around the kinds of block representations, avoiding the unclear "raw". (Any meaning of "raw" assumes some bias toward the storage layer or toward the logical data layer.) Preferred terminology: * **Serialized block** - bytes that go into storage. For block-based table (usually the case) this includes the block trailer. WART: block `size` may or may not include the trailer; need to be clear about whether it does or not. * **Maybe compressed block** - like a serialized block, but without the trailer (or no promise of including a trailer). Must be accompanied by a CompressionType. * **Uncompressed block** - "payload" bytes that are either stored with no compression, used as input to compression function, or result of decompression function. * **Parsed block** - an in-memory form of a block in block cache, as it is used by the table reader. Different C++ types are used depending on the block type (see block_like_traits.h). Other refactorings: * Misc corrections/improvements of internal API comments * Remove a few misleading / unhelpful / redundant comments. * Use move semantics in some places to simplify contracts * Use better parameter names to indicate which parameters are used for outputs * Remove some extraneous `extern` * Various clean-ups to `CacheDumperImpl` (mostly unnecessary code) Pull Request resolved: https://github.com/facebook/rocksdb/pull/10408 Test Plan: existing tests Reviewed By: akankshamahajan15 Differential Revision: D38172617 Pulled By: pdillinger fbshipit-source-id: ccb99299f324ac5ca46996d34c5089621a4f260c
2 years ago
"UncompressBlockData:"
"TamperWithDecompressionOutput",
[](void* arg) {
BlockContents* contents = static_cast<BlockContents*>(arg);
// Ensure uncompressed data != original data
const size_t len = contents->data.size() + 1;
std::unique_ptr<char[]> fake_data(new char[len]());
*contents = BlockContents(std::move(fake_data), len);
});
}
std::map<std::string, std::string> key_value_written;
const int kKeySize = 5;
const int kValUnitSize = 16;
const int kValSize = 256;
Random rnd(405);
Status s = Status::OK();
DestroyAndReopen(options);
// Write 10 random files
for (int i = 0; i < 10; i++) {
for (int j = 0; j < 5; j++) {
std::string key = rnd.RandomString(kKeySize);
// Ensure good compression ratio
std::string valueUnit = rnd.RandomString(kValUnitSize);
std::string value;
for (int k = 0; k < kValSize; k += kValUnitSize) {
value += valueUnit;
}
s = Put(key, value);
if (compression_failure_type_ == kTestCompressionFail) {
key_value_written[key] = value;
ASSERT_OK(s);
}
}
s = Flush();
if (compression_failure_type_ == kTestCompressionFail) {
ASSERT_OK(s);
}
s = dbfull()->TEST_WaitForCompact();
if (compression_failure_type_ == kTestCompressionFail) {
ASSERT_OK(s);
}
if (i == 4) {
// Make compression fail at the mid of table building
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
}
}
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
if (compression_failure_type_ == kTestCompressionFail) {
// Should be kNoCompression, check content consistency
std::unique_ptr<Iterator> db_iter(db_->NewIterator(ReadOptions()));
for (db_iter->SeekToFirst(); db_iter->Valid(); db_iter->Next()) {
std::string key = db_iter->key().ToString();
std::string value = db_iter->value().ToString();
ASSERT_NE(key_value_written.find(key), key_value_written.end());
ASSERT_EQ(key_value_written[key], value);
key_value_written.erase(key);
}
ASSERT_EQ(0, key_value_written.size());
} else if (compression_failure_type_ == kTestDecompressionFail) {
ASSERT_EQ(std::string(s.getState()),
"Could not decompress: kTestDecompressionFail");
} else if (compression_failure_type_ == kTestDecompressionCorruption) {
ASSERT_EQ(std::string(s.getState()),
Refactor to avoid confusing "raw block" (#10408) Summary: We have a lot of confusing code because of mixed, sometimes completely opposite uses of of the term "raw block" or "raw contents", sometimes within the same source file. For example, in `BlockBasedTableBuilder`, `raw_block_contents` and `raw_size` generally referred to uncompressed block contents and size, while `WriteRawBlock` referred to writing a block that is already compressed if it is going to be. Meanwhile, in `BlockBasedTable`, `raw_block_contents` either referred to a (maybe compressed) block with trailer, or a maybe compressed block maybe without trailer. (Note: left as follow-up work to use C++ typing to better sort out the various kinds of BlockContents.) This change primarily tries to apply some consistent terminology around the kinds of block representations, avoiding the unclear "raw". (Any meaning of "raw" assumes some bias toward the storage layer or toward the logical data layer.) Preferred terminology: * **Serialized block** - bytes that go into storage. For block-based table (usually the case) this includes the block trailer. WART: block `size` may or may not include the trailer; need to be clear about whether it does or not. * **Maybe compressed block** - like a serialized block, but without the trailer (or no promise of including a trailer). Must be accompanied by a CompressionType. * **Uncompressed block** - "payload" bytes that are either stored with no compression, used as input to compression function, or result of decompression function. * **Parsed block** - an in-memory form of a block in block cache, as it is used by the table reader. Different C++ types are used depending on the block type (see block_like_traits.h). Other refactorings: * Misc corrections/improvements of internal API comments * Remove a few misleading / unhelpful / redundant comments. * Use move semantics in some places to simplify contracts * Use better parameter names to indicate which parameters are used for outputs * Remove some extraneous `extern` * Various clean-ups to `CacheDumperImpl` (mostly unnecessary code) Pull Request resolved: https://github.com/facebook/rocksdb/pull/10408 Test Plan: existing tests Reviewed By: akankshamahajan15 Differential Revision: D38172617 Pulled By: pdillinger fbshipit-source-id: ccb99299f324ac5ca46996d34c5089621a4f260c
2 years ago
"Decompressed block did not match pre-compression block");
}
}
TEST_F(DBTest2, CompressionOptions) {
if (!Zlib_Supported() || !Snappy_Supported()) {
return;
}
Options options = CurrentOptions();
options.level0_file_num_compaction_trigger = 2;
options.max_bytes_for_level_base = 100;
options.max_bytes_for_level_multiplier = 2;
options.num_levels = 7;
options.max_background_compactions = 1;
CompactionCompressionListener* listener =
new CompactionCompressionListener(&options);
options.listeners.emplace_back(listener);
const int kKeySize = 5;
const int kValSize = 20;
Random rnd(301);
std::vector<uint32_t> compression_parallel_threads = {1, 4};
std::map<std::string, std::string> key_value_written;
for (int iter = 0; iter <= 2; iter++) {
listener->max_level_checked = 0;
if (iter == 0) {
// Use different compression algorithms for different levels but
// always use Zlib for bottommost level
options.compression_per_level = {kNoCompression, kNoCompression,
kNoCompression, kSnappyCompression,
kSnappyCompression, kSnappyCompression,
kZlibCompression};
options.compression = kNoCompression;
options.bottommost_compression = kZlibCompression;
} else if (iter == 1) {
// Use Snappy except for bottommost level use ZLib
options.compression_per_level = {};
options.compression = kSnappyCompression;
options.bottommost_compression = kZlibCompression;
} else if (iter == 2) {
// Use Snappy everywhere
options.compression_per_level = {};
options.compression = kSnappyCompression;
options.bottommost_compression = kDisableCompressionOption;
}
for (auto num_threads : compression_parallel_threads) {
options.compression_opts.parallel_threads = num_threads;
options.bottommost_compression_opts.parallel_threads = num_threads;
DestroyAndReopen(options);
// Write 10 random files
for (int i = 0; i < 10; i++) {
for (int j = 0; j < 5; j++) {
std::string key = rnd.RandomString(kKeySize);
std::string value = rnd.RandomString(kValSize);
key_value_written[key] = value;
ASSERT_OK(Put(key, value));
}
ASSERT_OK(Flush());
ASSERT_OK(dbfull()->TEST_WaitForCompact());
}
// Make sure that we wrote enough to check all 7 levels
ASSERT_EQ(listener->max_level_checked, 6);
// Make sure database content is the same as key_value_written
std::unique_ptr<Iterator> db_iter(db_->NewIterator(ReadOptions()));
for (db_iter->SeekToFirst(); db_iter->Valid(); db_iter->Next()) {
std::string key = db_iter->key().ToString();
std::string value = db_iter->value().ToString();
ASSERT_NE(key_value_written.find(key), key_value_written.end());
ASSERT_EQ(key_value_written[key], value);
key_value_written.erase(key);
}
ASSERT_OK(db_iter->status());
ASSERT_EQ(0, key_value_written.size());
}
}
}
class CompactionStallTestListener : public EventListener {
public:
CompactionStallTestListener() : compacting_files_cnt_(0), compacted_files_cnt_(0) {}
void OnCompactionBegin(DB* /*db*/, const CompactionJobInfo& ci) override {
ASSERT_EQ(ci.cf_name, "default");
ASSERT_EQ(ci.base_input_level, 0);
ASSERT_EQ(ci.compaction_reason, CompactionReason::kLevelL0FilesNum);
compacting_files_cnt_ += ci.input_files.size();
}
void OnCompactionCompleted(DB* /*db*/, const CompactionJobInfo& ci) override {
ASSERT_EQ(ci.cf_name, "default");
ASSERT_EQ(ci.base_input_level, 0);
ASSERT_EQ(ci.compaction_reason, CompactionReason::kLevelL0FilesNum);
compacted_files_cnt_ += ci.input_files.size();
}
std::atomic<size_t> compacting_files_cnt_;
std::atomic<size_t> compacted_files_cnt_;
};
TEST_F(DBTest2, CompactionStall) {
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->LoadDependency(
{{"DBImpl::BGWorkCompaction", "DBTest2::CompactionStall:0"},
{"DBImpl::BGWorkCompaction", "DBTest2::CompactionStall:1"},
{"DBTest2::CompactionStall:2",
"DBImpl::NotifyOnCompactionBegin::UnlockMutex"},
{"DBTest2::CompactionStall:3",
"DBImpl::NotifyOnCompactionCompleted::UnlockMutex"}});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
Options options = CurrentOptions();
options.level0_file_num_compaction_trigger = 4;
options.max_background_compactions = 40;
CompactionStallTestListener* listener = new CompactionStallTestListener();
options.listeners.emplace_back(listener);
DestroyAndReopen(options);
// make sure all background compaction jobs can be scheduled
auto stop_token =
dbfull()->TEST_write_controler().GetCompactionPressureToken();
Random rnd(301);
// 4 Files in L0
for (int i = 0; i < 4; i++) {
for (int j = 0; j < 10; j++) {
ASSERT_OK(Put(rnd.RandomString(10), rnd.RandomString(10)));
}
ASSERT_OK(Flush());
}
// Wait for compaction to be triggered
TEST_SYNC_POINT("DBTest2::CompactionStall:0");
// Clear "DBImpl::BGWorkCompaction" SYNC_POINT since we want to hold it again
// at DBTest2::CompactionStall::1
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearTrace();
// Another 6 L0 files to trigger compaction again
for (int i = 0; i < 6; i++) {
for (int j = 0; j < 10; j++) {
ASSERT_OK(Put(rnd.RandomString(10), rnd.RandomString(10)));
}
ASSERT_OK(Flush());
}
// Wait for another compaction to be triggered
TEST_SYNC_POINT("DBTest2::CompactionStall:1");
// Hold NotifyOnCompactionBegin in the unlock mutex section
TEST_SYNC_POINT("DBTest2::CompactionStall:2");
// Hold NotifyOnCompactionCompleted in the unlock mutex section
TEST_SYNC_POINT("DBTest2::CompactionStall:3");
ASSERT_OK(dbfull()->TEST_WaitForCompact());
ASSERT_LT(NumTableFilesAtLevel(0),
options.level0_file_num_compaction_trigger);
ASSERT_GT(listener->compacted_files_cnt_.load(),
10 - options.level0_file_num_compaction_trigger);
ASSERT_EQ(listener->compacting_files_cnt_.load(), listener->compacted_files_cnt_.load());
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
#endif // ROCKSDB_LITE
TEST_F(DBTest2, FirstSnapshotTest) {
Options options;
options.write_buffer_size = 100000; // Small write buffer
options = CurrentOptions(options);
CreateAndReopenWithCF({"pikachu"}, options);
// This snapshot will have sequence number 0 what is expected behaviour.
const Snapshot* s1 = db_->GetSnapshot();
ASSERT_OK(Put(1, "k1", std::string(100000, 'x'))); // Fill memtable
ASSERT_OK(Put(1, "k2", std::string(100000, 'y'))); // Trigger flush
db_->ReleaseSnapshot(s1);
}
#ifndef ROCKSDB_LITE
TEST_F(DBTest2, DuplicateSnapshot) {
Options options;
options = CurrentOptions(options);
std::vector<const Snapshot*> snapshots;
DBImpl* dbi = static_cast_with_check<DBImpl>(db_);
SequenceNumber oldest_ww_snap, first_ww_snap;
ASSERT_OK(Put("k", "v")); // inc seq
snapshots.push_back(db_->GetSnapshot());
snapshots.push_back(db_->GetSnapshot());
ASSERT_OK(Put("k", "v")); // inc seq
snapshots.push_back(db_->GetSnapshot());
snapshots.push_back(dbi->GetSnapshotForWriteConflictBoundary());
first_ww_snap = snapshots.back()->GetSequenceNumber();
ASSERT_OK(Put("k", "v")); // inc seq
snapshots.push_back(dbi->GetSnapshotForWriteConflictBoundary());
snapshots.push_back(db_->GetSnapshot());
ASSERT_OK(Put("k", "v")); // inc seq
snapshots.push_back(db_->GetSnapshot());
{
InstrumentedMutexLock l(dbi->mutex());
auto seqs = dbi->snapshots().GetAll(&oldest_ww_snap);
ASSERT_EQ(seqs.size(), 4); // duplicates are not counted
ASSERT_EQ(oldest_ww_snap, first_ww_snap);
}
for (auto s : snapshots) {
db_->ReleaseSnapshot(s);
}
}
#endif // ROCKSDB_LITE
class PinL0IndexAndFilterBlocksTest
: public DBTestBase,
public testing::WithParamInterface<std::tuple<bool, bool>> {
public:
PinL0IndexAndFilterBlocksTest()
: DBTestBase("db_pin_l0_index_bloom_test", /*env_do_fsync=*/true) {}
void SetUp() override {
infinite_max_files_ = std::get<0>(GetParam());
disallow_preload_ = std::get<1>(GetParam());
}
void CreateTwoLevels(Options* options, bool close_afterwards) {
if (infinite_max_files_) {
options->max_open_files = -1;
}
options->create_if_missing = true;
options->statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
BlockBasedTableOptions table_options;
table_options.cache_index_and_filter_blocks = true;
table_options.pin_l0_filter_and_index_blocks_in_cache = true;
table_options.filter_policy.reset(NewBloomFilterPolicy(20));
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
options->table_factory.reset(NewBlockBasedTableFactory(table_options));
CreateAndReopenWithCF({"pikachu"}, *options);
ASSERT_OK(Put(1, "a", "begin"));
ASSERT_OK(Put(1, "z", "end"));
ASSERT_OK(Flush(1));
// move this table to L1
ASSERT_OK(dbfull()->TEST_CompactRange(0, nullptr, nullptr, handles_[1]));
// reset block cache
table_options.block_cache = NewLRUCache(64 * 1024);
options->table_factory.reset(NewBlockBasedTableFactory(table_options));
TryReopenWithColumnFamilies({"default", "pikachu"}, *options);
// create new table at L0
ASSERT_OK(Put(1, "a2", "begin2"));
ASSERT_OK(Put(1, "z2", "end2"));
ASSERT_OK(Flush(1));
if (close_afterwards) {
Close(); // This ensures that there is no ref to block cache entries
}
table_options.block_cache->EraseUnRefEntries();
}
bool infinite_max_files_;
bool disallow_preload_;
};
TEST_P(PinL0IndexAndFilterBlocksTest,
IndexAndFilterBlocksOfNewTableAddedToCacheWithPinning) {
Options options = CurrentOptions();
if (infinite_max_files_) {
options.max_open_files = -1;
}
options.create_if_missing = true;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
BlockBasedTableOptions table_options;
table_options.cache_index_and_filter_blocks = true;
table_options.pin_l0_filter_and_index_blocks_in_cache = true;
table_options.filter_policy.reset(NewBloomFilterPolicy(20));
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
CreateAndReopenWithCF({"pikachu"}, options);
ASSERT_OK(Put(1, "key", "val"));
// Create a new table.
ASSERT_OK(Flush(1));
// index/filter blocks added to block cache right after table creation.
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
// only index/filter were added
ASSERT_EQ(2, TestGetTickerCount(options, BLOCK_CACHE_ADD));
ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_DATA_MISS));
std::string value;
// Miss and hit count should remain the same, they're all pinned.
ASSERT_TRUE(db_->KeyMayExist(ReadOptions(), handles_[1], "key", &value));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
// Miss and hit count should remain the same, they're all pinned.
value = Get(1, "key");
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
ASSERT_EQ(1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(0, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
}
TEST_P(PinL0IndexAndFilterBlocksTest,
MultiLevelIndexAndFilterBlocksCachedWithPinning) {
Options options = CurrentOptions();
PinL0IndexAndFilterBlocksTest::CreateTwoLevels(&options, false);
// get base cache values
uint64_t fm = TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS);
uint64_t fh = TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT);
uint64_t im = TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS);
uint64_t ih = TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT);
std::string value;
// this should be read from L0
// so cache values don't change
value = Get(1, "a2");
ASSERT_EQ(fm, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(fh, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
ASSERT_EQ(im, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(ih, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
// this should be read from L1
// the file is opened, prefetching results in a cache filter miss
// the block is loaded and added to the cache,
// then the get results in a cache hit for L1
// When we have inifinite max_files, there is still cache miss because we have
// reset the block cache
value = Get(1, "a");
ASSERT_EQ(fm + 1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(im + 1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
}
TEST_P(PinL0IndexAndFilterBlocksTest, DisablePrefetchingNonL0IndexAndFilter) {
Options options = CurrentOptions();
// This ensures that db does not ref anything in the block cache, so
// EraseUnRefEntries could clear them up.
bool close_afterwards = true;
PinL0IndexAndFilterBlocksTest::CreateTwoLevels(&options, close_afterwards);
// Get base cache values
uint64_t fm = TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS);
uint64_t fh = TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT);
uint64_t im = TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS);
uint64_t ih = TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT);
if (disallow_preload_) {
// Now we have two files. We narrow the max open files to allow 3 entries
// so that preloading SST files won't happen.
options.max_open_files = 13;
// RocksDB sanitize max open files to at least 20. Modify it back.
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"SanitizeOptions::AfterChangeMaxOpenFiles", [&](void* arg) {
int* max_open_files = static_cast<int*>(arg);
*max_open_files = 13;
});
}
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
// Reopen database. If max_open_files is set as -1, table readers will be
// preloaded. This will trigger a BlockBasedTable::Open() and prefetch
// L0 index and filter. Level 1's prefetching is disabled in DB::Open()
TryReopenWithColumnFamilies({"default", "pikachu"}, options);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
if (!disallow_preload_) {
// After reopen, cache miss are increased by one because we read (and only
// read) filter and index on L0
ASSERT_EQ(fm + 1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(fh, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
ASSERT_EQ(im + 1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(ih, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
} else {
// If max_open_files is not -1, we do not preload table readers, so there is
// no change.
ASSERT_EQ(fm, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(fh, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
ASSERT_EQ(im, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(ih, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
}
std::string value;
// this should be read from L0
value = Get(1, "a2");
// If max_open_files is -1, we have pinned index and filter in Rep, so there
// will not be changes in index and filter misses or hits. If max_open_files
// is not -1, Get() will open a TableReader and prefetch index and filter.
ASSERT_EQ(fm + 1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(fh, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
ASSERT_EQ(im + 1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(ih, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
// this should be read from L1
value = Get(1, "a");
if (!disallow_preload_) {
// In inifinite max files case, there's a cache miss in executing Get()
// because index and filter are not prefetched before.
ASSERT_EQ(fm + 2, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(fh, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
ASSERT_EQ(im + 2, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(ih, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
} else {
// In this case, cache miss will be increased by one in
// BlockBasedTable::Open() because this is not in DB::Open() code path so we
// will prefetch L1's index and filter. Cache hit will also be increased by
// one because Get() will read index and filter from the block cache
// prefetched in previous Open() call.
ASSERT_EQ(fm + 2, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(fh + 1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
ASSERT_EQ(im + 2, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(ih + 1, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
}
// Force a full compaction to one single file. There will be a block
// cache read for both of index and filter. If prefetch doesn't explicitly
// happen, it will happen when verifying the file.
Compact(1, "a", "zzzzz");
ASSERT_OK(dbfull()->TEST_WaitForCompact());
if (!disallow_preload_) {
ASSERT_EQ(fm + 3, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(fh, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
ASSERT_EQ(im + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(ih + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
} else {
ASSERT_EQ(fm + 3, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(fh + 1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
ASSERT_EQ(im + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(ih + 4, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
}
// Bloom and index hit will happen when a Get() happens.
value = Get(1, "a");
if (!disallow_preload_) {
ASSERT_EQ(fm + 3, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(fh + 1, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
ASSERT_EQ(im + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(ih + 4, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
} else {
ASSERT_EQ(fm + 3, TestGetTickerCount(options, BLOCK_CACHE_FILTER_MISS));
ASSERT_EQ(fh + 2, TestGetTickerCount(options, BLOCK_CACHE_FILTER_HIT));
ASSERT_EQ(im + 3, TestGetTickerCount(options, BLOCK_CACHE_INDEX_MISS));
ASSERT_EQ(ih + 5, TestGetTickerCount(options, BLOCK_CACHE_INDEX_HIT));
}
}
INSTANTIATE_TEST_CASE_P(PinL0IndexAndFilterBlocksTest,
PinL0IndexAndFilterBlocksTest,
::testing::Values(std::make_tuple(true, false),
std::make_tuple(false, false),
std::make_tuple(false, true)));
#ifndef ROCKSDB_LITE
TEST_F(DBTest2, MaxCompactionBytesTest) {
Options options = CurrentOptions();
options.memtable_factory.reset(test::NewSpecialSkipListFactory(
DBTestBase::kNumKeysByGenerateNewRandomFile));
options.compaction_style = kCompactionStyleLevel;
options.write_buffer_size = 200 << 10;
options.arena_block_size = 4 << 10;
options.level0_file_num_compaction_trigger = 4;
options.num_levels = 4;
options.compression = kNoCompression;
options.max_bytes_for_level_base = 450 << 10;
options.target_file_size_base = 100 << 10;
// Infinite for full compaction.
options.max_compaction_bytes = options.target_file_size_base * 100;
Reopen(options);
Random rnd(301);
for (int num = 0; num < 8; num++) {
GenerateNewRandomFile(&rnd);
}
CompactRangeOptions cro;
cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
ASSERT_EQ("0,0,8", FilesPerLevel(0));
// When compact from Ln -> Ln+1, cut a file if the file overlaps with
// more than three files in Ln+1.
options.max_compaction_bytes = options.target_file_size_base * 3;
Reopen(options);
GenerateNewRandomFile(&rnd);
// Add three more small files that overlap with the previous file
for (int i = 0; i < 3; i++) {
ASSERT_OK(Put("a", "z"));
ASSERT_OK(Flush());
}
ASSERT_OK(dbfull()->TEST_WaitForCompact());
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2 years ago
// Output files to L1 are cut to 4 pieces, according to
// options.max_compaction_bytes (300K)
// There are 8 files on L2 (grandparents level), each one is 100K. The first
// file overlaps with a, b which max_compaction_bytes is less than 300K, the
// second one overlaps with d, e, which is also less than 300K. Including any
// extra grandparent file will make the future compaction larger than 300K.
// L1: [ 1 ] [ 2 ] [ 3 ] [ 4 ]
// L2: [a] [b] [c] [d] [e] [f] [g] [h]
ASSERT_EQ("0,4,8", FilesPerLevel(0));
}
static void UniqueIdCallback(void* arg) {
int* result = reinterpret_cast<int*>(arg);
if (*result == -1) {
*result = 0;
}
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearTrace();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"GetUniqueIdFromFile:FS_IOC_GETVERSION", UniqueIdCallback);
}
class MockPersistentCache : public PersistentCache {
public:
explicit MockPersistentCache(const bool is_compressed, const size_t max_size)
: is_compressed_(is_compressed), max_size_(max_size) {
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"GetUniqueIdFromFile:FS_IOC_GETVERSION", UniqueIdCallback);
}
~MockPersistentCache() override {}
PersistentCache::StatsType Stats() override {
return PersistentCache::StatsType();
}
uint64_t NewId() override {
return last_id_.fetch_add(1, std::memory_order_relaxed);
}
Status Insert(const Slice& page_key, const char* data,
const size_t size) override {
MutexLock _(&lock_);
if (size_ > max_size_) {
size_ -= data_.begin()->second.size();
data_.erase(data_.begin());
}
data_.insert(std::make_pair(page_key.ToString(), std::string(data, size)));
size_ += size;
return Status::OK();
}
Status Lookup(const Slice& page_key, std::unique_ptr<char[]>* data,
size_t* size) override {
MutexLock _(&lock_);
auto it = data_.find(page_key.ToString());
if (it == data_.end()) {
return Status::NotFound();
}
assert(page_key.ToString() == it->first);
data->reset(new char[it->second.size()]);
memcpy(data->get(), it->second.c_str(), it->second.size());
*size = it->second.size();
return Status::OK();
}
bool IsCompressed() override { return is_compressed_; }
std::string GetPrintableOptions() const override {
return "MockPersistentCache";
}
port::Mutex lock_;
std::map<std::string, std::string> data_;
const bool is_compressed_ = true;
size_t size_ = 0;
const size_t max_size_ = 10 * 1024; // 10KiB
std::atomic<uint64_t> last_id_{1};
};
#ifdef OS_LINUX
// Make sure that in CPU time perf context counters, Env::NowCPUNanos()
// is used, rather than Env::CPUNanos();
TEST_F(DBTest2, TestPerfContextGetCpuTime) {
// force resizing table cache so table handle is not preloaded so that
// we can measure find_table_nanos during Get().
dbfull()->TEST_table_cache()->SetCapacity(0);
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Flush());
env_->now_cpu_count_.store(0);
Fix+clean up handling of mock sleeps (#7101) Summary: We have a number of tests hanging on MacOS and windows due to mishandling of code for mock sleeps. In addition, the code was in terrible shape because the same variable (addon_time_) would sometimes refer to microseconds and sometimes to seconds. One test even assumed it was nanoseconds but was written to pass anyway. This has been cleaned up so that DB tests generally use a SpecialEnv function to mock sleep, for either some number of microseconds or seconds depending on the function called. But to call one of these, the test must first call SetMockSleep (precondition enforced with assertion), which also turns sleeps in RocksDB into mock sleeps. To also removes accounting for actual clock time, call SetTimeElapseOnlySleepOnReopen, which implies SetMockSleep (on DB re-open). This latter setting only works by applying on DB re-open, otherwise havoc can ensue if Env goes back in time with DB open. More specifics: Removed some unused test classes, and updated comments on the general problem. Fixed DBSSTTest.GetTotalSstFilesSize using a sync point callback instead of mock time. For this we have the only modification to production code, inserting a sync point callback in flush_job.cc, which is not a change to production behavior. Removed unnecessary resetting of mock times to 0 in many tests. RocksDB deals in relative time. Any behaviors relying on absolute date/time are likely a bug. (The above test DBSSTTest.GetTotalSstFilesSize was the only one clearly injecting a specific absolute time for actual testing convenience.) Just in case I misunderstood some test, I put this note in each replacement: // NOTE: Presumed unnecessary and removed: resetting mock time in env Strengthened some tests like MergeTestTime, MergeCompactionTimeTest, and FilterCompactionTimeTest in db_test.cc stats_history_test and blob_db_test are each their own beast, rather deeply dependent on MockTimeEnv. Each gets its own variant of a work-around for TimedWait in a mock time environment. (Reduces redundancy and inconsistency in stats_history_test.) Intended follow-up: Remove TimedWait from the public API of InstrumentedCondVar, and only make that accessible through Env by passing in an InstrumentedCondVar and a deadline. Then the Env implementations mocking time can fix this problem without using sync points. (Test infrastructure using sync points interferes with individual tests' control over sync points.) With that change, we can simplify/consolidate the scattered work-arounds. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7101 Test Plan: make check on Linux and MacOS Reviewed By: zhichao-cao Differential Revision: D23032815 Pulled By: pdillinger fbshipit-source-id: 7f33967ada8b83011fb54e8279365c008bd6610b
4 years ago
env_->SetMockSleep();
// NOTE: Presumed unnecessary and removed: resetting mock time in env
// CPU timing is not enabled with kEnableTimeExceptForMutex
SetPerfLevel(PerfLevel::kEnableTimeExceptForMutex);
ASSERT_EQ("bar", Get("foo"));
ASSERT_EQ(0, get_perf_context()->get_cpu_nanos);
ASSERT_EQ(0, env_->now_cpu_count_.load());
Fix+clean up handling of mock sleeps (#7101) Summary: We have a number of tests hanging on MacOS and windows due to mishandling of code for mock sleeps. In addition, the code was in terrible shape because the same variable (addon_time_) would sometimes refer to microseconds and sometimes to seconds. One test even assumed it was nanoseconds but was written to pass anyway. This has been cleaned up so that DB tests generally use a SpecialEnv function to mock sleep, for either some number of microseconds or seconds depending on the function called. But to call one of these, the test must first call SetMockSleep (precondition enforced with assertion), which also turns sleeps in RocksDB into mock sleeps. To also removes accounting for actual clock time, call SetTimeElapseOnlySleepOnReopen, which implies SetMockSleep (on DB re-open). This latter setting only works by applying on DB re-open, otherwise havoc can ensue if Env goes back in time with DB open. More specifics: Removed some unused test classes, and updated comments on the general problem. Fixed DBSSTTest.GetTotalSstFilesSize using a sync point callback instead of mock time. For this we have the only modification to production code, inserting a sync point callback in flush_job.cc, which is not a change to production behavior. Removed unnecessary resetting of mock times to 0 in many tests. RocksDB deals in relative time. Any behaviors relying on absolute date/time are likely a bug. (The above test DBSSTTest.GetTotalSstFilesSize was the only one clearly injecting a specific absolute time for actual testing convenience.) Just in case I misunderstood some test, I put this note in each replacement: // NOTE: Presumed unnecessary and removed: resetting mock time in env Strengthened some tests like MergeTestTime, MergeCompactionTimeTest, and FilterCompactionTimeTest in db_test.cc stats_history_test and blob_db_test are each their own beast, rather deeply dependent on MockTimeEnv. Each gets its own variant of a work-around for TimedWait in a mock time environment. (Reduces redundancy and inconsistency in stats_history_test.) Intended follow-up: Remove TimedWait from the public API of InstrumentedCondVar, and only make that accessible through Env by passing in an InstrumentedCondVar and a deadline. Then the Env implementations mocking time can fix this problem without using sync points. (Test infrastructure using sync points interferes with individual tests' control over sync points.) With that change, we can simplify/consolidate the scattered work-arounds. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7101 Test Plan: make check on Linux and MacOS Reviewed By: zhichao-cao Differential Revision: D23032815 Pulled By: pdillinger fbshipit-source-id: 7f33967ada8b83011fb54e8279365c008bd6610b
4 years ago
constexpr uint64_t kDummyAddonSeconds = uint64_t{1000000};
constexpr uint64_t kDummyAddonNanos = 1000000000U * kDummyAddonSeconds;
// Add time to NowNanos() reading.
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"TableCache::FindTable:0",
Fix+clean up handling of mock sleeps (#7101) Summary: We have a number of tests hanging on MacOS and windows due to mishandling of code for mock sleeps. In addition, the code was in terrible shape because the same variable (addon_time_) would sometimes refer to microseconds and sometimes to seconds. One test even assumed it was nanoseconds but was written to pass anyway. This has been cleaned up so that DB tests generally use a SpecialEnv function to mock sleep, for either some number of microseconds or seconds depending on the function called. But to call one of these, the test must first call SetMockSleep (precondition enforced with assertion), which also turns sleeps in RocksDB into mock sleeps. To also removes accounting for actual clock time, call SetTimeElapseOnlySleepOnReopen, which implies SetMockSleep (on DB re-open). This latter setting only works by applying on DB re-open, otherwise havoc can ensue if Env goes back in time with DB open. More specifics: Removed some unused test classes, and updated comments on the general problem. Fixed DBSSTTest.GetTotalSstFilesSize using a sync point callback instead of mock time. For this we have the only modification to production code, inserting a sync point callback in flush_job.cc, which is not a change to production behavior. Removed unnecessary resetting of mock times to 0 in many tests. RocksDB deals in relative time. Any behaviors relying on absolute date/time are likely a bug. (The above test DBSSTTest.GetTotalSstFilesSize was the only one clearly injecting a specific absolute time for actual testing convenience.) Just in case I misunderstood some test, I put this note in each replacement: // NOTE: Presumed unnecessary and removed: resetting mock time in env Strengthened some tests like MergeTestTime, MergeCompactionTimeTest, and FilterCompactionTimeTest in db_test.cc stats_history_test and blob_db_test are each their own beast, rather deeply dependent on MockTimeEnv. Each gets its own variant of a work-around for TimedWait in a mock time environment. (Reduces redundancy and inconsistency in stats_history_test.) Intended follow-up: Remove TimedWait from the public API of InstrumentedCondVar, and only make that accessible through Env by passing in an InstrumentedCondVar and a deadline. Then the Env implementations mocking time can fix this problem without using sync points. (Test infrastructure using sync points interferes with individual tests' control over sync points.) With that change, we can simplify/consolidate the scattered work-arounds. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7101 Test Plan: make check on Linux and MacOS Reviewed By: zhichao-cao Differential Revision: D23032815 Pulled By: pdillinger fbshipit-source-id: 7f33967ada8b83011fb54e8279365c008bd6610b
4 years ago
[&](void* /*arg*/) { env_->MockSleepForSeconds(kDummyAddonSeconds); });
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
SetPerfLevel(PerfLevel::kEnableTimeAndCPUTimeExceptForMutex);
ASSERT_EQ("bar", Get("foo"));
ASSERT_GT(env_->now_cpu_count_.load(), 2);
Fix+clean up handling of mock sleeps (#7101) Summary: We have a number of tests hanging on MacOS and windows due to mishandling of code for mock sleeps. In addition, the code was in terrible shape because the same variable (addon_time_) would sometimes refer to microseconds and sometimes to seconds. One test even assumed it was nanoseconds but was written to pass anyway. This has been cleaned up so that DB tests generally use a SpecialEnv function to mock sleep, for either some number of microseconds or seconds depending on the function called. But to call one of these, the test must first call SetMockSleep (precondition enforced with assertion), which also turns sleeps in RocksDB into mock sleeps. To also removes accounting for actual clock time, call SetTimeElapseOnlySleepOnReopen, which implies SetMockSleep (on DB re-open). This latter setting only works by applying on DB re-open, otherwise havoc can ensue if Env goes back in time with DB open. More specifics: Removed some unused test classes, and updated comments on the general problem. Fixed DBSSTTest.GetTotalSstFilesSize using a sync point callback instead of mock time. For this we have the only modification to production code, inserting a sync point callback in flush_job.cc, which is not a change to production behavior. Removed unnecessary resetting of mock times to 0 in many tests. RocksDB deals in relative time. Any behaviors relying on absolute date/time are likely a bug. (The above test DBSSTTest.GetTotalSstFilesSize was the only one clearly injecting a specific absolute time for actual testing convenience.) Just in case I misunderstood some test, I put this note in each replacement: // NOTE: Presumed unnecessary and removed: resetting mock time in env Strengthened some tests like MergeTestTime, MergeCompactionTimeTest, and FilterCompactionTimeTest in db_test.cc stats_history_test and blob_db_test are each their own beast, rather deeply dependent on MockTimeEnv. Each gets its own variant of a work-around for TimedWait in a mock time environment. (Reduces redundancy and inconsistency in stats_history_test.) Intended follow-up: Remove TimedWait from the public API of InstrumentedCondVar, and only make that accessible through Env by passing in an InstrumentedCondVar and a deadline. Then the Env implementations mocking time can fix this problem without using sync points. (Test infrastructure using sync points interferes with individual tests' control over sync points.) With that change, we can simplify/consolidate the scattered work-arounds. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7101 Test Plan: make check on Linux and MacOS Reviewed By: zhichao-cao Differential Revision: D23032815 Pulled By: pdillinger fbshipit-source-id: 7f33967ada8b83011fb54e8279365c008bd6610b
4 years ago
ASSERT_LT(get_perf_context()->get_cpu_nanos, kDummyAddonNanos);
ASSERT_GT(get_perf_context()->find_table_nanos, kDummyAddonNanos);
SetPerfLevel(PerfLevel::kDisable);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, TestPerfContextIterCpuTime) {
DestroyAndReopen(CurrentOptions());
// force resizing table cache so table handle is not preloaded so that
// we can measure find_table_nanos during iteration
dbfull()->TEST_table_cache()->SetCapacity(0);
const size_t kNumEntries = 10;
for (size_t i = 0; i < kNumEntries; ++i) {
ASSERT_OK(Put("k" + std::to_string(i), "v" + std::to_string(i)));
}
ASSERT_OK(Flush());
for (size_t i = 0; i < kNumEntries; ++i) {
ASSERT_EQ("v" + std::to_string(i), Get("k" + std::to_string(i)));
}
std::string last_key = "k" + std::to_string(kNumEntries - 1);
std::string last_value = "v" + std::to_string(kNumEntries - 1);
env_->now_cpu_count_.store(0);
Fix+clean up handling of mock sleeps (#7101) Summary: We have a number of tests hanging on MacOS and windows due to mishandling of code for mock sleeps. In addition, the code was in terrible shape because the same variable (addon_time_) would sometimes refer to microseconds and sometimes to seconds. One test even assumed it was nanoseconds but was written to pass anyway. This has been cleaned up so that DB tests generally use a SpecialEnv function to mock sleep, for either some number of microseconds or seconds depending on the function called. But to call one of these, the test must first call SetMockSleep (precondition enforced with assertion), which also turns sleeps in RocksDB into mock sleeps. To also removes accounting for actual clock time, call SetTimeElapseOnlySleepOnReopen, which implies SetMockSleep (on DB re-open). This latter setting only works by applying on DB re-open, otherwise havoc can ensue if Env goes back in time with DB open. More specifics: Removed some unused test classes, and updated comments on the general problem. Fixed DBSSTTest.GetTotalSstFilesSize using a sync point callback instead of mock time. For this we have the only modification to production code, inserting a sync point callback in flush_job.cc, which is not a change to production behavior. Removed unnecessary resetting of mock times to 0 in many tests. RocksDB deals in relative time. Any behaviors relying on absolute date/time are likely a bug. (The above test DBSSTTest.GetTotalSstFilesSize was the only one clearly injecting a specific absolute time for actual testing convenience.) Just in case I misunderstood some test, I put this note in each replacement: // NOTE: Presumed unnecessary and removed: resetting mock time in env Strengthened some tests like MergeTestTime, MergeCompactionTimeTest, and FilterCompactionTimeTest in db_test.cc stats_history_test and blob_db_test are each their own beast, rather deeply dependent on MockTimeEnv. Each gets its own variant of a work-around for TimedWait in a mock time environment. (Reduces redundancy and inconsistency in stats_history_test.) Intended follow-up: Remove TimedWait from the public API of InstrumentedCondVar, and only make that accessible through Env by passing in an InstrumentedCondVar and a deadline. Then the Env implementations mocking time can fix this problem without using sync points. (Test infrastructure using sync points interferes with individual tests' control over sync points.) With that change, we can simplify/consolidate the scattered work-arounds. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7101 Test Plan: make check on Linux and MacOS Reviewed By: zhichao-cao Differential Revision: D23032815 Pulled By: pdillinger fbshipit-source-id: 7f33967ada8b83011fb54e8279365c008bd6610b
4 years ago
env_->SetMockSleep();
// NOTE: Presumed unnecessary and removed: resetting mock time in env
// CPU timing is not enabled with kEnableTimeExceptForMutex
SetPerfLevel(PerfLevel::kEnableTimeExceptForMutex);
Iterator* iter = db_->NewIterator(ReadOptions());
iter->Seek("k0");
ASSERT_TRUE(iter->Valid());
ASSERT_EQ("v0", iter->value().ToString());
iter->SeekForPrev(last_key);
ASSERT_TRUE(iter->Valid());
iter->SeekToLast();
ASSERT_TRUE(iter->Valid());
ASSERT_EQ(last_value, iter->value().ToString());
iter->SeekToFirst();
ASSERT_TRUE(iter->Valid());
ASSERT_EQ("v0", iter->value().ToString());
ASSERT_EQ(0, get_perf_context()->iter_seek_cpu_nanos);
iter->Next();
ASSERT_TRUE(iter->Valid());
ASSERT_EQ("v1", iter->value().ToString());
ASSERT_EQ(0, get_perf_context()->iter_next_cpu_nanos);
iter->Prev();
ASSERT_TRUE(iter->Valid());
ASSERT_OK(iter->status());
ASSERT_EQ("v0", iter->value().ToString());
ASSERT_EQ(0, get_perf_context()->iter_prev_cpu_nanos);
ASSERT_EQ(0, env_->now_cpu_count_.load());
delete iter;
Fix+clean up handling of mock sleeps (#7101) Summary: We have a number of tests hanging on MacOS and windows due to mishandling of code for mock sleeps. In addition, the code was in terrible shape because the same variable (addon_time_) would sometimes refer to microseconds and sometimes to seconds. One test even assumed it was nanoseconds but was written to pass anyway. This has been cleaned up so that DB tests generally use a SpecialEnv function to mock sleep, for either some number of microseconds or seconds depending on the function called. But to call one of these, the test must first call SetMockSleep (precondition enforced with assertion), which also turns sleeps in RocksDB into mock sleeps. To also removes accounting for actual clock time, call SetTimeElapseOnlySleepOnReopen, which implies SetMockSleep (on DB re-open). This latter setting only works by applying on DB re-open, otherwise havoc can ensue if Env goes back in time with DB open. More specifics: Removed some unused test classes, and updated comments on the general problem. Fixed DBSSTTest.GetTotalSstFilesSize using a sync point callback instead of mock time. For this we have the only modification to production code, inserting a sync point callback in flush_job.cc, which is not a change to production behavior. Removed unnecessary resetting of mock times to 0 in many tests. RocksDB deals in relative time. Any behaviors relying on absolute date/time are likely a bug. (The above test DBSSTTest.GetTotalSstFilesSize was the only one clearly injecting a specific absolute time for actual testing convenience.) Just in case I misunderstood some test, I put this note in each replacement: // NOTE: Presumed unnecessary and removed: resetting mock time in env Strengthened some tests like MergeTestTime, MergeCompactionTimeTest, and FilterCompactionTimeTest in db_test.cc stats_history_test and blob_db_test are each their own beast, rather deeply dependent on MockTimeEnv. Each gets its own variant of a work-around for TimedWait in a mock time environment. (Reduces redundancy and inconsistency in stats_history_test.) Intended follow-up: Remove TimedWait from the public API of InstrumentedCondVar, and only make that accessible through Env by passing in an InstrumentedCondVar and a deadline. Then the Env implementations mocking time can fix this problem without using sync points. (Test infrastructure using sync points interferes with individual tests' control over sync points.) With that change, we can simplify/consolidate the scattered work-arounds. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7101 Test Plan: make check on Linux and MacOS Reviewed By: zhichao-cao Differential Revision: D23032815 Pulled By: pdillinger fbshipit-source-id: 7f33967ada8b83011fb54e8279365c008bd6610b
4 years ago
constexpr uint64_t kDummyAddonSeconds = uint64_t{1000000};
constexpr uint64_t kDummyAddonNanos = 1000000000U * kDummyAddonSeconds;
// Add time to NowNanos() reading.
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"TableCache::FindTable:0",
Fix+clean up handling of mock sleeps (#7101) Summary: We have a number of tests hanging on MacOS and windows due to mishandling of code for mock sleeps. In addition, the code was in terrible shape because the same variable (addon_time_) would sometimes refer to microseconds and sometimes to seconds. One test even assumed it was nanoseconds but was written to pass anyway. This has been cleaned up so that DB tests generally use a SpecialEnv function to mock sleep, for either some number of microseconds or seconds depending on the function called. But to call one of these, the test must first call SetMockSleep (precondition enforced with assertion), which also turns sleeps in RocksDB into mock sleeps. To also removes accounting for actual clock time, call SetTimeElapseOnlySleepOnReopen, which implies SetMockSleep (on DB re-open). This latter setting only works by applying on DB re-open, otherwise havoc can ensue if Env goes back in time with DB open. More specifics: Removed some unused test classes, and updated comments on the general problem. Fixed DBSSTTest.GetTotalSstFilesSize using a sync point callback instead of mock time. For this we have the only modification to production code, inserting a sync point callback in flush_job.cc, which is not a change to production behavior. Removed unnecessary resetting of mock times to 0 in many tests. RocksDB deals in relative time. Any behaviors relying on absolute date/time are likely a bug. (The above test DBSSTTest.GetTotalSstFilesSize was the only one clearly injecting a specific absolute time for actual testing convenience.) Just in case I misunderstood some test, I put this note in each replacement: // NOTE: Presumed unnecessary and removed: resetting mock time in env Strengthened some tests like MergeTestTime, MergeCompactionTimeTest, and FilterCompactionTimeTest in db_test.cc stats_history_test and blob_db_test are each their own beast, rather deeply dependent on MockTimeEnv. Each gets its own variant of a work-around for TimedWait in a mock time environment. (Reduces redundancy and inconsistency in stats_history_test.) Intended follow-up: Remove TimedWait from the public API of InstrumentedCondVar, and only make that accessible through Env by passing in an InstrumentedCondVar and a deadline. Then the Env implementations mocking time can fix this problem without using sync points. (Test infrastructure using sync points interferes with individual tests' control over sync points.) With that change, we can simplify/consolidate the scattered work-arounds. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7101 Test Plan: make check on Linux and MacOS Reviewed By: zhichao-cao Differential Revision: D23032815 Pulled By: pdillinger fbshipit-source-id: 7f33967ada8b83011fb54e8279365c008bd6610b
4 years ago
[&](void* /*arg*/) { env_->MockSleepForSeconds(kDummyAddonSeconds); });
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
SetPerfLevel(PerfLevel::kEnableTimeAndCPUTimeExceptForMutex);
iter = db_->NewIterator(ReadOptions());
iter->Seek("k0");
ASSERT_TRUE(iter->Valid());
ASSERT_EQ("v0", iter->value().ToString());
iter->SeekForPrev(last_key);
ASSERT_TRUE(iter->Valid());
iter->SeekToLast();
ASSERT_TRUE(iter->Valid());
ASSERT_EQ(last_value, iter->value().ToString());
iter->SeekToFirst();
ASSERT_TRUE(iter->Valid());
ASSERT_EQ("v0", iter->value().ToString());
ASSERT_GT(get_perf_context()->iter_seek_cpu_nanos, 0);
Fix+clean up handling of mock sleeps (#7101) Summary: We have a number of tests hanging on MacOS and windows due to mishandling of code for mock sleeps. In addition, the code was in terrible shape because the same variable (addon_time_) would sometimes refer to microseconds and sometimes to seconds. One test even assumed it was nanoseconds but was written to pass anyway. This has been cleaned up so that DB tests generally use a SpecialEnv function to mock sleep, for either some number of microseconds or seconds depending on the function called. But to call one of these, the test must first call SetMockSleep (precondition enforced with assertion), which also turns sleeps in RocksDB into mock sleeps. To also removes accounting for actual clock time, call SetTimeElapseOnlySleepOnReopen, which implies SetMockSleep (on DB re-open). This latter setting only works by applying on DB re-open, otherwise havoc can ensue if Env goes back in time with DB open. More specifics: Removed some unused test classes, and updated comments on the general problem. Fixed DBSSTTest.GetTotalSstFilesSize using a sync point callback instead of mock time. For this we have the only modification to production code, inserting a sync point callback in flush_job.cc, which is not a change to production behavior. Removed unnecessary resetting of mock times to 0 in many tests. RocksDB deals in relative time. Any behaviors relying on absolute date/time are likely a bug. (The above test DBSSTTest.GetTotalSstFilesSize was the only one clearly injecting a specific absolute time for actual testing convenience.) Just in case I misunderstood some test, I put this note in each replacement: // NOTE: Presumed unnecessary and removed: resetting mock time in env Strengthened some tests like MergeTestTime, MergeCompactionTimeTest, and FilterCompactionTimeTest in db_test.cc stats_history_test and blob_db_test are each their own beast, rather deeply dependent on MockTimeEnv. Each gets its own variant of a work-around for TimedWait in a mock time environment. (Reduces redundancy and inconsistency in stats_history_test.) Intended follow-up: Remove TimedWait from the public API of InstrumentedCondVar, and only make that accessible through Env by passing in an InstrumentedCondVar and a deadline. Then the Env implementations mocking time can fix this problem without using sync points. (Test infrastructure using sync points interferes with individual tests' control over sync points.) With that change, we can simplify/consolidate the scattered work-arounds. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7101 Test Plan: make check on Linux and MacOS Reviewed By: zhichao-cao Differential Revision: D23032815 Pulled By: pdillinger fbshipit-source-id: 7f33967ada8b83011fb54e8279365c008bd6610b
4 years ago
ASSERT_LT(get_perf_context()->iter_seek_cpu_nanos, kDummyAddonNanos);
iter->Next();
ASSERT_TRUE(iter->Valid());
ASSERT_EQ("v1", iter->value().ToString());
ASSERT_GT(get_perf_context()->iter_next_cpu_nanos, 0);
Fix+clean up handling of mock sleeps (#7101) Summary: We have a number of tests hanging on MacOS and windows due to mishandling of code for mock sleeps. In addition, the code was in terrible shape because the same variable (addon_time_) would sometimes refer to microseconds and sometimes to seconds. One test even assumed it was nanoseconds but was written to pass anyway. This has been cleaned up so that DB tests generally use a SpecialEnv function to mock sleep, for either some number of microseconds or seconds depending on the function called. But to call one of these, the test must first call SetMockSleep (precondition enforced with assertion), which also turns sleeps in RocksDB into mock sleeps. To also removes accounting for actual clock time, call SetTimeElapseOnlySleepOnReopen, which implies SetMockSleep (on DB re-open). This latter setting only works by applying on DB re-open, otherwise havoc can ensue if Env goes back in time with DB open. More specifics: Removed some unused test classes, and updated comments on the general problem. Fixed DBSSTTest.GetTotalSstFilesSize using a sync point callback instead of mock time. For this we have the only modification to production code, inserting a sync point callback in flush_job.cc, which is not a change to production behavior. Removed unnecessary resetting of mock times to 0 in many tests. RocksDB deals in relative time. Any behaviors relying on absolute date/time are likely a bug. (The above test DBSSTTest.GetTotalSstFilesSize was the only one clearly injecting a specific absolute time for actual testing convenience.) Just in case I misunderstood some test, I put this note in each replacement: // NOTE: Presumed unnecessary and removed: resetting mock time in env Strengthened some tests like MergeTestTime, MergeCompactionTimeTest, and FilterCompactionTimeTest in db_test.cc stats_history_test and blob_db_test are each their own beast, rather deeply dependent on MockTimeEnv. Each gets its own variant of a work-around for TimedWait in a mock time environment. (Reduces redundancy and inconsistency in stats_history_test.) Intended follow-up: Remove TimedWait from the public API of InstrumentedCondVar, and only make that accessible through Env by passing in an InstrumentedCondVar and a deadline. Then the Env implementations mocking time can fix this problem without using sync points. (Test infrastructure using sync points interferes with individual tests' control over sync points.) With that change, we can simplify/consolidate the scattered work-arounds. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7101 Test Plan: make check on Linux and MacOS Reviewed By: zhichao-cao Differential Revision: D23032815 Pulled By: pdillinger fbshipit-source-id: 7f33967ada8b83011fb54e8279365c008bd6610b
4 years ago
ASSERT_LT(get_perf_context()->iter_next_cpu_nanos, kDummyAddonNanos);
iter->Prev();
ASSERT_TRUE(iter->Valid());
ASSERT_OK(iter->status());
ASSERT_EQ("v0", iter->value().ToString());
ASSERT_GT(get_perf_context()->iter_prev_cpu_nanos, 0);
Fix+clean up handling of mock sleeps (#7101) Summary: We have a number of tests hanging on MacOS and windows due to mishandling of code for mock sleeps. In addition, the code was in terrible shape because the same variable (addon_time_) would sometimes refer to microseconds and sometimes to seconds. One test even assumed it was nanoseconds but was written to pass anyway. This has been cleaned up so that DB tests generally use a SpecialEnv function to mock sleep, for either some number of microseconds or seconds depending on the function called. But to call one of these, the test must first call SetMockSleep (precondition enforced with assertion), which also turns sleeps in RocksDB into mock sleeps. To also removes accounting for actual clock time, call SetTimeElapseOnlySleepOnReopen, which implies SetMockSleep (on DB re-open). This latter setting only works by applying on DB re-open, otherwise havoc can ensue if Env goes back in time with DB open. More specifics: Removed some unused test classes, and updated comments on the general problem. Fixed DBSSTTest.GetTotalSstFilesSize using a sync point callback instead of mock time. For this we have the only modification to production code, inserting a sync point callback in flush_job.cc, which is not a change to production behavior. Removed unnecessary resetting of mock times to 0 in many tests. RocksDB deals in relative time. Any behaviors relying on absolute date/time are likely a bug. (The above test DBSSTTest.GetTotalSstFilesSize was the only one clearly injecting a specific absolute time for actual testing convenience.) Just in case I misunderstood some test, I put this note in each replacement: // NOTE: Presumed unnecessary and removed: resetting mock time in env Strengthened some tests like MergeTestTime, MergeCompactionTimeTest, and FilterCompactionTimeTest in db_test.cc stats_history_test and blob_db_test are each their own beast, rather deeply dependent on MockTimeEnv. Each gets its own variant of a work-around for TimedWait in a mock time environment. (Reduces redundancy and inconsistency in stats_history_test.) Intended follow-up: Remove TimedWait from the public API of InstrumentedCondVar, and only make that accessible through Env by passing in an InstrumentedCondVar and a deadline. Then the Env implementations mocking time can fix this problem without using sync points. (Test infrastructure using sync points interferes with individual tests' control over sync points.) With that change, we can simplify/consolidate the scattered work-arounds. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7101 Test Plan: make check on Linux and MacOS Reviewed By: zhichao-cao Differential Revision: D23032815 Pulled By: pdillinger fbshipit-source-id: 7f33967ada8b83011fb54e8279365c008bd6610b
4 years ago
ASSERT_LT(get_perf_context()->iter_prev_cpu_nanos, kDummyAddonNanos);
ASSERT_GE(env_->now_cpu_count_.load(), 12);
Fix+clean up handling of mock sleeps (#7101) Summary: We have a number of tests hanging on MacOS and windows due to mishandling of code for mock sleeps. In addition, the code was in terrible shape because the same variable (addon_time_) would sometimes refer to microseconds and sometimes to seconds. One test even assumed it was nanoseconds but was written to pass anyway. This has been cleaned up so that DB tests generally use a SpecialEnv function to mock sleep, for either some number of microseconds or seconds depending on the function called. But to call one of these, the test must first call SetMockSleep (precondition enforced with assertion), which also turns sleeps in RocksDB into mock sleeps. To also removes accounting for actual clock time, call SetTimeElapseOnlySleepOnReopen, which implies SetMockSleep (on DB re-open). This latter setting only works by applying on DB re-open, otherwise havoc can ensue if Env goes back in time with DB open. More specifics: Removed some unused test classes, and updated comments on the general problem. Fixed DBSSTTest.GetTotalSstFilesSize using a sync point callback instead of mock time. For this we have the only modification to production code, inserting a sync point callback in flush_job.cc, which is not a change to production behavior. Removed unnecessary resetting of mock times to 0 in many tests. RocksDB deals in relative time. Any behaviors relying on absolute date/time are likely a bug. (The above test DBSSTTest.GetTotalSstFilesSize was the only one clearly injecting a specific absolute time for actual testing convenience.) Just in case I misunderstood some test, I put this note in each replacement: // NOTE: Presumed unnecessary and removed: resetting mock time in env Strengthened some tests like MergeTestTime, MergeCompactionTimeTest, and FilterCompactionTimeTest in db_test.cc stats_history_test and blob_db_test are each their own beast, rather deeply dependent on MockTimeEnv. Each gets its own variant of a work-around for TimedWait in a mock time environment. (Reduces redundancy and inconsistency in stats_history_test.) Intended follow-up: Remove TimedWait from the public API of InstrumentedCondVar, and only make that accessible through Env by passing in an InstrumentedCondVar and a deadline. Then the Env implementations mocking time can fix this problem without using sync points. (Test infrastructure using sync points interferes with individual tests' control over sync points.) With that change, we can simplify/consolidate the scattered work-arounds. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7101 Test Plan: make check on Linux and MacOS Reviewed By: zhichao-cao Differential Revision: D23032815 Pulled By: pdillinger fbshipit-source-id: 7f33967ada8b83011fb54e8279365c008bd6610b
4 years ago
ASSERT_GT(get_perf_context()->find_table_nanos, kDummyAddonNanos);
SetPerfLevel(PerfLevel::kDisable);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
delete iter;
}
#endif // OS_LINUX
#if !defined OS_SOLARIS
TEST_F(DBTest2, PersistentCache) {
int num_iter = 80;
Options options;
options.write_buffer_size = 64 * 1024; // small write buffer
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
options = CurrentOptions(options);
auto bsizes = {/*no block cache*/ 0, /*1M*/ 1 * 1024 * 1024};
auto types = {/*compressed*/ 1, /*uncompressed*/ 0};
for (auto bsize : bsizes) {
for (auto type : types) {
BlockBasedTableOptions table_options;
table_options.persistent_cache.reset(
new MockPersistentCache(type, 10 * 1024));
table_options.no_block_cache = true;
table_options.block_cache = bsize ? NewLRUCache(bsize) : nullptr;
table_options.block_cache_compressed = nullptr;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
DestroyAndReopen(options);
CreateAndReopenWithCF({"pikachu"}, options);
// default column family doesn't have block cache
Options no_block_cache_opts;
no_block_cache_opts.statistics = options.statistics;
no_block_cache_opts = CurrentOptions(no_block_cache_opts);
BlockBasedTableOptions table_options_no_bc;
table_options_no_bc.no_block_cache = true;
no_block_cache_opts.table_factory.reset(
NewBlockBasedTableFactory(table_options_no_bc));
ReopenWithColumnFamilies(
{"default", "pikachu"},
std::vector<Options>({no_block_cache_opts, options}));
Random rnd(301);
// Write 8MB (80 values, each 100K)
ASSERT_EQ(NumTableFilesAtLevel(0, 1), 0);
std::vector<std::string> values;
std::string str;
for (int i = 0; i < num_iter; i++) {
if (i % 4 == 0) { // high compression ratio
str = rnd.RandomString(1000);
}
values.push_back(str);
ASSERT_OK(Put(1, Key(i), values[i]));
}
// flush all data from memtable so that reads are from block cache
ASSERT_OK(Flush(1));
for (int i = 0; i < num_iter; i++) {
ASSERT_EQ(Get(1, Key(i)), values[i]);
}
auto hit = options.statistics->getTickerCount(PERSISTENT_CACHE_HIT);
auto miss = options.statistics->getTickerCount(PERSISTENT_CACHE_MISS);
ASSERT_GT(hit, 0);
ASSERT_GT(miss, 0);
}
}
}
#endif // !defined OS_SOLARIS
namespace {
void CountSyncPoint() {
TEST_SYNC_POINT_CALLBACK("DBTest2::MarkedPoint", nullptr /* arg */);
}
} // namespace
TEST_F(DBTest2, SyncPointMarker) {
std::atomic<int> sync_point_called(0);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"DBTest2::MarkedPoint",
[&](void* /*arg*/) { sync_point_called.fetch_add(1); });
// The first dependency enforces Marker can be loaded before MarkedPoint.
// The second checks that thread 1's MarkedPoint should be disabled here.
// Execution order:
// | Thread 1 | Thread 2 |
// | | Marker |
// | MarkedPoint | |
// | Thread1First | |
// | | MarkedPoint |
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->LoadDependencyAndMarkers(
{{"DBTest2::SyncPointMarker:Thread1First", "DBTest2::MarkedPoint"}},
{{"DBTest2::SyncPointMarker:Marker", "DBTest2::MarkedPoint"}});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
std::function<void()> func1 = [&]() {
CountSyncPoint();
TEST_SYNC_POINT("DBTest2::SyncPointMarker:Thread1First");
};
std::function<void()> func2 = [&]() {
TEST_SYNC_POINT("DBTest2::SyncPointMarker:Marker");
CountSyncPoint();
};
auto thread1 = port::Thread(func1);
auto thread2 = port::Thread(func2);
thread1.join();
thread2.join();
// Callback is only executed once
ASSERT_EQ(sync_point_called.load(), 1);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
#endif
Introduce FullMergeV2 (eliminate memcpy from merge operators) Summary: This diff update the code to pin the merge operator operands while the merge operation is done, so that we can eliminate the memcpy cost, to do that we need a new public API for FullMerge that replace the std::deque<std::string> with std::vector<Slice> This diff is stacked on top of D56493 and D56511 In this diff we - Update FullMergeV2 arguments to be encapsulated in MergeOperationInput and MergeOperationOutput which will make it easier to add new arguments in the future - Replace std::deque<std::string> with std::vector<Slice> to pass operands - Replace MergeContext std::deque with std::vector (based on a simple benchmark I ran https://gist.github.com/IslamAbdelRahman/78fc86c9ab9f52b1df791e58943fb187) - Allow FullMergeV2 output to be an existing operand ``` [Everything in Memtable | 10K operands | 10 KB each | 1 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=10000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 0.607 micros/op 1648235 ops/sec; 16121.2 MB/s readseq : 0.478 micros/op 2091546 ops/sec; 20457.2 MB/s readseq : 0.252 micros/op 3972081 ops/sec; 38850.5 MB/s readseq : 0.237 micros/op 4218328 ops/sec; 41259.0 MB/s readseq : 0.247 micros/op 4043927 ops/sec; 39553.2 MB/s [master] readseq : 3.935 micros/op 254140 ops/sec; 2485.7 MB/s readseq : 3.722 micros/op 268657 ops/sec; 2627.7 MB/s readseq : 3.149 micros/op 317605 ops/sec; 3106.5 MB/s readseq : 3.125 micros/op 320024 ops/sec; 3130.1 MB/s readseq : 4.075 micros/op 245374 ops/sec; 2400.0 MB/s ``` ``` [Everything in Memtable | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="mergerandom,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=1000 --num=10000 --disable_auto_compactions --value_size=10240 --write_buffer_size=1000000000 [FullMergeV2] readseq : 3.472 micros/op 288018 ops/sec; 2817.1 MB/s readseq : 2.304 micros/op 434027 ops/sec; 4245.2 MB/s readseq : 1.163 micros/op 859845 ops/sec; 8410.0 MB/s readseq : 1.192 micros/op 838926 ops/sec; 8205.4 MB/s readseq : 1.250 micros/op 800000 ops/sec; 7824.7 MB/s [master] readseq : 24.025 micros/op 41623 ops/sec; 407.1 MB/s readseq : 18.489 micros/op 54086 ops/sec; 529.0 MB/s readseq : 18.693 micros/op 53495 ops/sec; 523.2 MB/s readseq : 23.621 micros/op 42335 ops/sec; 414.1 MB/s readseq : 18.775 micros/op 53262 ops/sec; 521.0 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 1 operand per key] [FullMergeV2] $ DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions readseq : 14.741 micros/op 67837 ops/sec; 663.5 MB/s readseq : 1.029 micros/op 971446 ops/sec; 9501.6 MB/s readseq : 0.974 micros/op 1026229 ops/sec; 10037.4 MB/s readseq : 0.965 micros/op 1036080 ops/sec; 10133.8 MB/s readseq : 0.943 micros/op 1060657 ops/sec; 10374.2 MB/s [master] readseq : 16.735 micros/op 59755 ops/sec; 584.5 MB/s readseq : 3.029 micros/op 330151 ops/sec; 3229.2 MB/s readseq : 3.136 micros/op 318883 ops/sec; 3119.0 MB/s readseq : 3.065 micros/op 326245 ops/sec; 3191.0 MB/s readseq : 3.014 micros/op 331813 ops/sec; 3245.4 MB/s ``` ``` [Everything in Block cache | 10K operands | 10 KB each | 10 operand per key] DEBUG_LEVEL=0 make db_bench -j64 && ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --num=100000 --db="/dev/shm/merge-random-10-operands-10K-10KB" --cache_size=1000000000 --use_existing_db --disable_auto_compactions [FullMergeV2] readseq : 24.325 micros/op 41109 ops/sec; 402.1 MB/s readseq : 1.470 micros/op 680272 ops/sec; 6653.7 MB/s readseq : 1.231 micros/op 812347 ops/sec; 7945.5 MB/s readseq : 1.091 micros/op 916590 ops/sec; 8965.1 MB/s readseq : 1.109 micros/op 901713 ops/sec; 8819.6 MB/s [master] readseq : 27.257 micros/op 36687 ops/sec; 358.8 MB/s readseq : 4.443 micros/op 225073 ops/sec; 2201.4 MB/s readseq : 5.830 micros/op 171526 ops/sec; 1677.7 MB/s readseq : 4.173 micros/op 239635 ops/sec; 2343.8 MB/s readseq : 4.150 micros/op 240963 ops/sec; 2356.8 MB/s ``` Test Plan: COMPILE_WITH_ASAN=1 make check -j64 Reviewers: yhchiang, andrewkr, sdong Reviewed By: sdong Subscribers: lovro, andrewkr, dhruba Differential Revision: https://reviews.facebook.net/D57075
8 years ago
size_t GetEncodedEntrySize(size_t key_size, size_t value_size) {
std::string buffer;
PutVarint32(&buffer, static_cast<uint32_t>(0));
PutVarint32(&buffer, static_cast<uint32_t>(key_size));
PutVarint32(&buffer, static_cast<uint32_t>(value_size));
return buffer.size() + key_size + value_size;
}
TEST_F(DBTest2, ReadAmpBitmap) {
Options options = CurrentOptions();
BlockBasedTableOptions bbto;
uint32_t bytes_per_bit[2] = {1, 16};
for (size_t k = 0; k < 2; k++) {
// Disable delta encoding to make it easier to calculate read amplification
bbto.use_delta_encoding = false;
// Huge block cache to make it easier to calculate read amplification
bbto.block_cache = NewLRUCache(1024 * 1024 * 1024);
bbto.read_amp_bytes_per_bit = bytes_per_bit[k];
options.table_factory.reset(NewBlockBasedTableFactory(bbto));
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
DestroyAndReopen(options);
const size_t kNumEntries = 10000;
Random rnd(301);
for (size_t i = 0; i < kNumEntries; i++) {
ASSERT_OK(Put(Key(static_cast<int>(i)), rnd.RandomString(100)));
}
ASSERT_OK(Flush());
Close();
Reopen(options);
// Read keys/values randomly and verify that reported read amp error
// is less than 2%
uint64_t total_useful_bytes = 0;
std::set<int> read_keys;
std::string value;
for (size_t i = 0; i < kNumEntries * 5; i++) {
int key_idx = rnd.Next() % kNumEntries;
std::string key = Key(key_idx);
ASSERT_OK(db_->Get(ReadOptions(), key, &value));
if (read_keys.find(key_idx) == read_keys.end()) {
auto internal_key = InternalKey(key, 0, ValueType::kTypeValue);
total_useful_bytes +=
GetEncodedEntrySize(internal_key.size(), value.size());
read_keys.insert(key_idx);
}
double expected_read_amp =
static_cast<double>(total_useful_bytes) /
options.statistics->getTickerCount(READ_AMP_TOTAL_READ_BYTES);
double read_amp =
static_cast<double>(options.statistics->getTickerCount(
READ_AMP_ESTIMATE_USEFUL_BYTES)) /
options.statistics->getTickerCount(READ_AMP_TOTAL_READ_BYTES);
double error_pct = fabs(expected_read_amp - read_amp) * 100;
// Error between reported read amp and real read amp should be less than
// 2%
EXPECT_LE(error_pct, 2);
}
// Make sure we read every thing in the DB (which is smaller than our cache)
Iterator* iter = db_->NewIterator(ReadOptions());
for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
ASSERT_EQ(iter->value().ToString(), Get(iter->key().ToString()));
}
ASSERT_OK(iter->status());
delete iter;
// Read amp is on average 100% since we read all what we loaded in memory
if (k == 0) {
ASSERT_EQ(
options.statistics->getTickerCount(READ_AMP_ESTIMATE_USEFUL_BYTES),
options.statistics->getTickerCount(READ_AMP_TOTAL_READ_BYTES));
} else {
ASSERT_NEAR(
options.statistics->getTickerCount(READ_AMP_ESTIMATE_USEFUL_BYTES) *
1.0f /
options.statistics->getTickerCount(READ_AMP_TOTAL_READ_BYTES),
1, .01);
}
}
}
#ifndef OS_SOLARIS // GetUniqueIdFromFile is not implemented
TEST_F(DBTest2, ReadAmpBitmapLiveInCacheAfterDBClose) {
{
const int kIdBufLen = 100;
char id_buf[kIdBufLen];
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
Status s = Status::NotSupported();
#ifndef OS_WIN
// You can't open a directory on windows using random access file
std::unique_ptr<RandomAccessFile> file;
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
s = env_->NewRandomAccessFile(dbname_, &file, EnvOptions());
if (s.ok()) {
if (file->GetUniqueId(id_buf, kIdBufLen) == 0) {
// fs holding db directory doesn't support getting a unique file id,
// this means that running this test will fail because lru_cache will
// load the blocks again regardless of them being already in the cache
return;
}
}
#endif
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
if (!s.ok()) {
std::unique_ptr<Directory> dir;
ASSERT_OK(env_->NewDirectory(dbname_, &dir));
if (dir->GetUniqueId(id_buf, kIdBufLen) == 0) {
// fs holding db directory doesn't support getting a unique file id,
// this means that running this test will fail because lru_cache will
// load the blocks again regardless of them being already in the cache
return;
}
}
}
uint32_t bytes_per_bit[2] = {1, 16};
for (size_t k = 0; k < 2; k++) {
std::shared_ptr<Cache> lru_cache = NewLRUCache(1024 * 1024 * 1024);
std::shared_ptr<Statistics> stats = ROCKSDB_NAMESPACE::CreateDBStatistics();
Options options = CurrentOptions();
BlockBasedTableOptions bbto;
// Disable delta encoding to make it easier to calculate read amplification
bbto.use_delta_encoding = false;
// Huge block cache to make it easier to calculate read amplification
bbto.block_cache = lru_cache;
bbto.read_amp_bytes_per_bit = bytes_per_bit[k];
options.table_factory.reset(NewBlockBasedTableFactory(bbto));
options.statistics = stats;
DestroyAndReopen(options);
const int kNumEntries = 10000;
Random rnd(301);
for (int i = 0; i < kNumEntries; i++) {
ASSERT_OK(Put(Key(i), rnd.RandomString(100)));
}
ASSERT_OK(Flush());
Close();
Reopen(options);
std::set<int> read_keys;
std::string value;
// Iter1: Read half the DB, Read even keys
// Key(0), Key(2), Key(4), Key(6), Key(8), ...
for (int i = 0; i < kNumEntries; i += 2) {
std::string key = Key(i);
ASSERT_OK(db_->Get(ReadOptions(), key, &value));
if (read_keys.find(i) == read_keys.end()) {
auto internal_key = InternalKey(key, 0, ValueType::kTypeValue);
read_keys.insert(i);
}
}
size_t total_useful_bytes_iter1 =
options.statistics->getTickerCount(READ_AMP_ESTIMATE_USEFUL_BYTES);
size_t total_loaded_bytes_iter1 =
options.statistics->getTickerCount(READ_AMP_TOTAL_READ_BYTES);
Close();
std::shared_ptr<Statistics> new_statistics =
ROCKSDB_NAMESPACE::CreateDBStatistics();
// Destroy old statistics obj that the blocks in lru_cache are pointing to
options.statistics.reset();
// Use the statistics object that we just created
options.statistics = new_statistics;
Reopen(options);
// Iter2: Read half the DB, Read odd keys
// Key(1), Key(3), Key(5), Key(7), Key(9), ...
for (int i = 1; i < kNumEntries; i += 2) {
std::string key = Key(i);
ASSERT_OK(db_->Get(ReadOptions(), key, &value));
if (read_keys.find(i) == read_keys.end()) {
auto internal_key = InternalKey(key, 0, ValueType::kTypeValue);
read_keys.insert(i);
}
}
size_t total_useful_bytes_iter2 =
options.statistics->getTickerCount(READ_AMP_ESTIMATE_USEFUL_BYTES);
size_t total_loaded_bytes_iter2 =
options.statistics->getTickerCount(READ_AMP_TOTAL_READ_BYTES);
// Read amp is on average 100% since we read all what we loaded in memory
if (k == 0) {
ASSERT_EQ(total_useful_bytes_iter1 + total_useful_bytes_iter2,
total_loaded_bytes_iter1 + total_loaded_bytes_iter2);
} else {
ASSERT_NEAR((total_useful_bytes_iter1 + total_useful_bytes_iter2) * 1.0f /
(total_loaded_bytes_iter1 + total_loaded_bytes_iter2),
1, .01);
}
}
}
#endif // !OS_SOLARIS
#ifndef ROCKSDB_LITE
TEST_F(DBTest2, AutomaticCompactionOverlapManualCompaction) {
Options options = CurrentOptions();
options.num_levels = 3;
options.IncreaseParallelism(20);
DestroyAndReopen(options);
ASSERT_OK(Put(Key(0), "a"));
ASSERT_OK(Put(Key(5), "a"));
ASSERT_OK(Flush());
ASSERT_OK(Put(Key(10), "a"));
ASSERT_OK(Put(Key(15), "a"));
ASSERT_OK(Flush());
CompactRangeOptions cro;
cro.change_level = true;
cro.target_level = 2;
ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
auto get_stat = [](std::string level_str, LevelStatType type,
std::map<std::string, std::string> props) {
auto prop_str =
"compaction." + level_str + "." +
InternalStats::compaction_level_stats.at(type).property_name.c_str();
auto prop_item = props.find(prop_str);
return prop_item == props.end() ? 0 : std::stod(prop_item->second);
};
// Trivial move 2 files to L2
ASSERT_EQ("0,0,2", FilesPerLevel());
// Also test that the stats GetMapProperty API reporting the same result
{
std::map<std::string, std::string> prop;
ASSERT_TRUE(dbfull()->GetMapProperty("rocksdb.cfstats", &prop));
ASSERT_EQ(0, get_stat("L0", LevelStatType::NUM_FILES, prop));
ASSERT_EQ(0, get_stat("L1", LevelStatType::NUM_FILES, prop));
ASSERT_EQ(2, get_stat("L2", LevelStatType::NUM_FILES, prop));
ASSERT_EQ(2, get_stat("Sum", LevelStatType::NUM_FILES, prop));
}
// While the compaction is running, we will create 2 new files that
// can fit in L2, these 2 files will be moved to L2 and overlap with
// the running compaction and break the LSM consistency.
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"CompactionJob::Run():Start", [&](void* /*arg*/) {
ASSERT_OK(
dbfull()->SetOptions({{"level0_file_num_compaction_trigger", "2"},
{"max_bytes_for_level_base", "1"}}));
ASSERT_OK(Put(Key(6), "a"));
ASSERT_OK(Put(Key(7), "a"));
ASSERT_OK(Flush());
ASSERT_OK(Put(Key(8), "a"));
ASSERT_OK(Put(Key(9), "a"));
ASSERT_OK(Flush());
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
// Run a manual compaction that will compact the 2 files in L2
// into 1 file in L2
cro.exclusive_manual_compaction = false;
cro.bottommost_level_compaction = BottommostLevelCompaction::kForceOptimized;
ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
// Test that the stats GetMapProperty API reporting 1 file in L2
{
std::map<std::string, std::string> prop;
ASSERT_TRUE(dbfull()->GetMapProperty("rocksdb.cfstats", &prop));
ASSERT_EQ(1, get_stat("L2", LevelStatType::NUM_FILES, prop));
}
}
TEST_F(DBTest2, ManualCompactionOverlapManualCompaction) {
Options options = CurrentOptions();
options.num_levels = 2;
options.IncreaseParallelism(20);
options.disable_auto_compactions = true;
DestroyAndReopen(options);
ASSERT_OK(Put(Key(0), "a"));
ASSERT_OK(Put(Key(5), "a"));
ASSERT_OK(Flush());
ASSERT_OK(Put(Key(10), "a"));
ASSERT_OK(Put(Key(15), "a"));
ASSERT_OK(Flush());
ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
// Trivial move 2 files to L1
ASSERT_EQ("0,2", FilesPerLevel());
std::function<void()> bg_manual_compact = [&]() {
std::string k1 = Key(6);
std::string k2 = Key(9);
Slice k1s(k1);
Slice k2s(k2);
CompactRangeOptions cro;
cro.exclusive_manual_compaction = false;
ASSERT_OK(db_->CompactRange(cro, &k1s, &k2s));
};
ROCKSDB_NAMESPACE::port::Thread bg_thread;
// While the compaction is running, we will create 2 new files that
// can fit in L1, these 2 files will be moved to L1 and overlap with
// the running compaction and break the LSM consistency.
std::atomic<bool> flag(false);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"CompactionJob::Run():Start", [&](void* /*arg*/) {
if (flag.exchange(true)) {
// We want to make sure to call this callback only once
return;
}
ASSERT_OK(Put(Key(6), "a"));
ASSERT_OK(Put(Key(7), "a"));
ASSERT_OK(Flush());
ASSERT_OK(Put(Key(8), "a"));
ASSERT_OK(Put(Key(9), "a"));
ASSERT_OK(Flush());
// Start a non-exclusive manual compaction in a bg thread
bg_thread = port::Thread(bg_manual_compact);
// This manual compaction conflict with the other manual compaction
// so it should wait until the first compaction finish
env_->SleepForMicroseconds(1000000);
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
// Run a manual compaction that will compact the 2 files in L1
// into 1 file in L1
CompactRangeOptions cro;
cro.exclusive_manual_compaction = false;
cro.bottommost_level_compaction = BottommostLevelCompaction::kForceOptimized;
ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
bg_thread.join();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, PausingManualCompaction1) {
Options options = CurrentOptions();
options.disable_auto_compactions = true;
options.num_levels = 7;
DestroyAndReopen(options);
Random rnd(301);
// Generate a file containing 10 keys.
for (int i = 0; i < 10; i++) {
ASSERT_OK(Put(Key(i), rnd.RandomString(50)));
}
ASSERT_OK(Flush());
// Generate another file containing same keys
for (int i = 0; i < 10; i++) {
ASSERT_OK(Put(Key(i), rnd.RandomString(50)));
}
ASSERT_OK(Flush());
int manual_compactions_paused = 0;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"CompactionJob::Run():PausingManualCompaction:1", [&](void* arg) {
auto canceled = static_cast<std::atomic<bool>*>(arg);
// CompactRange triggers manual compaction and cancel the compaction
// by set *canceled as true
if (canceled != nullptr) {
canceled->store(true, std::memory_order_release);
}
manual_compactions_paused += 1;
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"TestCompactFiles:PausingManualCompaction:3", [&](void* arg) {
Disable manual compaction during `ReFitLevel()` (#7250) Summary: Manual compaction with `CompactRangeOptions::change_levels` set could refit to a level targeted by another manual compaction. If force_consistency_checks were disabled, it could be possible for overlapping files to be written at that target level. This PR prevents the possibility by calling `DisableManualCompaction()` prior to `ReFitLevel()`. It also improves the manual compaction disabling mechanism to wait for pending manual compactions to complete before returning, and support disabling from multiple threads. Fixes https://github.com/facebook/rocksdb/issues/6432. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7250 Test Plan: crash test command that repro'd the bug reliably: ``` $ TEST_TMPDIR=/dev/shm python tools/db_crashtest.py blackbox --simple -target_file_size_base=524288 -write_buffer_size=1048576 -clear_column_family_one_in=0 -reopen=0 -max_key=10000000 -column_families=1 -max_background_compactions=8 -compact_range_one_in=100000 -compression_type=none -compaction_style=1 -num_levels=5 -universal_min_merge_width=4 -universal_max_merge_width=8 -level0_file_num_compaction_trigger=12 -rate_limiter_bytes_per_sec=1048576000 -universal_max_size_amplification_percent=100 --duration=3600 --interval=60 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --enable_compaction_filter=0 ``` Reviewed By: ltamasi Differential Revision: D23090800 Pulled By: ajkr fbshipit-source-id: afcbcd51b42ce76789fdb907d8b9ada790709c13
4 years ago
auto paused = static_cast<std::atomic<int>*>(arg);
// CompactFiles() relies on manual_compactions_paused to
// determine if thie compaction should be paused or not
Disable manual compaction during `ReFitLevel()` (#7250) Summary: Manual compaction with `CompactRangeOptions::change_levels` set could refit to a level targeted by another manual compaction. If force_consistency_checks were disabled, it could be possible for overlapping files to be written at that target level. This PR prevents the possibility by calling `DisableManualCompaction()` prior to `ReFitLevel()`. It also improves the manual compaction disabling mechanism to wait for pending manual compactions to complete before returning, and support disabling from multiple threads. Fixes https://github.com/facebook/rocksdb/issues/6432. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7250 Test Plan: crash test command that repro'd the bug reliably: ``` $ TEST_TMPDIR=/dev/shm python tools/db_crashtest.py blackbox --simple -target_file_size_base=524288 -write_buffer_size=1048576 -clear_column_family_one_in=0 -reopen=0 -max_key=10000000 -column_families=1 -max_background_compactions=8 -compact_range_one_in=100000 -compression_type=none -compaction_style=1 -num_levels=5 -universal_min_merge_width=4 -universal_max_merge_width=8 -level0_file_num_compaction_trigger=12 -rate_limiter_bytes_per_sec=1048576000 -universal_max_size_amplification_percent=100 --duration=3600 --interval=60 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --enable_compaction_filter=0 ``` Reviewed By: ltamasi Differential Revision: D23090800 Pulled By: ajkr fbshipit-source-id: afcbcd51b42ce76789fdb907d8b9ada790709c13
4 years ago
ASSERT_EQ(0, paused->load(std::memory_order_acquire));
paused->fetch_add(1, std::memory_order_release);
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
std::vector<std::string> files_before_compact, files_after_compact;
// Remember file name before compaction is triggered
std::vector<LiveFileMetaData> files_meta;
dbfull()->GetLiveFilesMetaData(&files_meta);
for (auto file : files_meta) {
files_before_compact.push_back(file.name);
}
// OK, now trigger a manual compaction
ASSERT_TRUE(dbfull()
->CompactRange(CompactRangeOptions(), nullptr, nullptr)
.IsManualCompactionPaused());
// Wait for compactions to get scheduled and stopped
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
// Get file names after compaction is stopped
files_meta.clear();
dbfull()->GetLiveFilesMetaData(&files_meta);
for (auto file : files_meta) {
files_after_compact.push_back(file.name);
}
// Like nothing happened
ASSERT_EQ(files_before_compact, files_after_compact);
ASSERT_EQ(manual_compactions_paused, 1);
manual_compactions_paused = 0;
// Now make sure CompactFiles also not run
ASSERT_TRUE(dbfull()
->CompactFiles(ROCKSDB_NAMESPACE::CompactionOptions(),
files_before_compact, 0)
.IsManualCompactionPaused());
// Wait for manual compaction to get scheduled and finish
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
files_meta.clear();
files_after_compact.clear();
dbfull()->GetLiveFilesMetaData(&files_meta);
for (auto file : files_meta) {
files_after_compact.push_back(file.name);
}
ASSERT_EQ(files_before_compact, files_after_compact);
// CompactFiles returns at entry point
ASSERT_EQ(manual_compactions_paused, 0);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
// PausingManualCompaction does not affect auto compaction
TEST_F(DBTest2, PausingManualCompaction2) {
Options options = CurrentOptions();
options.level0_file_num_compaction_trigger = 2;
options.disable_auto_compactions = false;
DestroyAndReopen(options);
dbfull()->DisableManualCompaction();
Random rnd(301);
for (int i = 0; i < 2; i++) {
// Generate a file containing 100 keys.
for (int j = 0; j < 100; j++) {
ASSERT_OK(Put(Key(j), rnd.RandomString(50)));
}
ASSERT_OK(Flush());
}
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
std::vector<LiveFileMetaData> files_meta;
dbfull()->GetLiveFilesMetaData(&files_meta);
ASSERT_EQ(files_meta.size(), 1);
}
TEST_F(DBTest2, PausingManualCompaction3) {
CompactRangeOptions compact_options;
Options options = CurrentOptions();
options.disable_auto_compactions = true;
options.num_levels = 7;
Random rnd(301);
auto generate_files = [&]() {
for (int i = 0; i < options.num_levels; i++) {
for (int j = 0; j < options.num_levels - i + 1; j++) {
for (int k = 0; k < 1000; k++) {
ASSERT_OK(Put(Key(k + j * 1000), rnd.RandomString(50)));
}
ASSERT_OK(Flush());
}
for (int l = 1; l < options.num_levels - i; l++) {
MoveFilesToLevel(l);
}
}
};
DestroyAndReopen(options);
generate_files();
#ifndef ROCKSDB_LITE
ASSERT_EQ("2,3,4,5,6,7,8", FilesPerLevel());
#endif // !ROCKSDB_LITE
int run_manual_compactions = 0;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"CompactionJob::Run():PausingManualCompaction:1",
[&](void* /*arg*/) { run_manual_compactions++; });
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
dbfull()->DisableManualCompaction();
ASSERT_TRUE(dbfull()
->CompactRange(compact_options, nullptr, nullptr)
.IsManualCompactionPaused());
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
// As manual compaction disabled, not even reach sync point
ASSERT_EQ(run_manual_compactions, 0);
#ifndef ROCKSDB_LITE
ASSERT_EQ("2,3,4,5,6,7,8", FilesPerLevel());
#endif // !ROCKSDB_LITE
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearCallBack(
"CompactionJob::Run():PausingManualCompaction:1");
dbfull()->EnableManualCompaction();
ASSERT_OK(dbfull()->CompactRange(compact_options, nullptr, nullptr));
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
#ifndef ROCKSDB_LITE
ASSERT_EQ("0,0,0,0,0,0,2", FilesPerLevel());
#endif // !ROCKSDB_LITE
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, PausingManualCompaction4) {
CompactRangeOptions compact_options;
Options options = CurrentOptions();
options.disable_auto_compactions = true;
options.num_levels = 7;
Random rnd(301);
auto generate_files = [&]() {
for (int i = 0; i < options.num_levels; i++) {
for (int j = 0; j < options.num_levels - i + 1; j++) {
for (int k = 0; k < 1000; k++) {
ASSERT_OK(Put(Key(k + j * 1000), rnd.RandomString(50)));
}
ASSERT_OK(Flush());
}
for (int l = 1; l < options.num_levels - i; l++) {
MoveFilesToLevel(l);
}
}
};
DestroyAndReopen(options);
generate_files();
#ifndef ROCKSDB_LITE
ASSERT_EQ("2,3,4,5,6,7,8", FilesPerLevel());
#endif // !ROCKSDB_LITE
int run_manual_compactions = 0;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"CompactionJob::Run():PausingManualCompaction:2", [&](void* arg) {
auto canceled = static_cast<std::atomic<bool>*>(arg);
// CompactRange triggers manual compaction and cancel the compaction
// by set *canceled as true
if (canceled != nullptr) {
canceled->store(true, std::memory_order_release);
}
run_manual_compactions++;
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"TestCompactFiles:PausingManualCompaction:3", [&](void* arg) {
Disable manual compaction during `ReFitLevel()` (#7250) Summary: Manual compaction with `CompactRangeOptions::change_levels` set could refit to a level targeted by another manual compaction. If force_consistency_checks were disabled, it could be possible for overlapping files to be written at that target level. This PR prevents the possibility by calling `DisableManualCompaction()` prior to `ReFitLevel()`. It also improves the manual compaction disabling mechanism to wait for pending manual compactions to complete before returning, and support disabling from multiple threads. Fixes https://github.com/facebook/rocksdb/issues/6432. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7250 Test Plan: crash test command that repro'd the bug reliably: ``` $ TEST_TMPDIR=/dev/shm python tools/db_crashtest.py blackbox --simple -target_file_size_base=524288 -write_buffer_size=1048576 -clear_column_family_one_in=0 -reopen=0 -max_key=10000000 -column_families=1 -max_background_compactions=8 -compact_range_one_in=100000 -compression_type=none -compaction_style=1 -num_levels=5 -universal_min_merge_width=4 -universal_max_merge_width=8 -level0_file_num_compaction_trigger=12 -rate_limiter_bytes_per_sec=1048576000 -universal_max_size_amplification_percent=100 --duration=3600 --interval=60 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --enable_compaction_filter=0 ``` Reviewed By: ltamasi Differential Revision: D23090800 Pulled By: ajkr fbshipit-source-id: afcbcd51b42ce76789fdb907d8b9ada790709c13
4 years ago
auto paused = static_cast<std::atomic<int>*>(arg);
// CompactFiles() relies on manual_compactions_paused to
// determine if thie compaction should be paused or not
Disable manual compaction during `ReFitLevel()` (#7250) Summary: Manual compaction with `CompactRangeOptions::change_levels` set could refit to a level targeted by another manual compaction. If force_consistency_checks were disabled, it could be possible for overlapping files to be written at that target level. This PR prevents the possibility by calling `DisableManualCompaction()` prior to `ReFitLevel()`. It also improves the manual compaction disabling mechanism to wait for pending manual compactions to complete before returning, and support disabling from multiple threads. Fixes https://github.com/facebook/rocksdb/issues/6432. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7250 Test Plan: crash test command that repro'd the bug reliably: ``` $ TEST_TMPDIR=/dev/shm python tools/db_crashtest.py blackbox --simple -target_file_size_base=524288 -write_buffer_size=1048576 -clear_column_family_one_in=0 -reopen=0 -max_key=10000000 -column_families=1 -max_background_compactions=8 -compact_range_one_in=100000 -compression_type=none -compaction_style=1 -num_levels=5 -universal_min_merge_width=4 -universal_max_merge_width=8 -level0_file_num_compaction_trigger=12 -rate_limiter_bytes_per_sec=1048576000 -universal_max_size_amplification_percent=100 --duration=3600 --interval=60 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --enable_compaction_filter=0 ``` Reviewed By: ltamasi Differential Revision: D23090800 Pulled By: ajkr fbshipit-source-id: afcbcd51b42ce76789fdb907d8b9ada790709c13
4 years ago
ASSERT_EQ(0, paused->load(std::memory_order_acquire));
paused->fetch_add(1, std::memory_order_release);
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
ASSERT_TRUE(dbfull()
->CompactRange(compact_options, nullptr, nullptr)
.IsManualCompactionPaused());
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
ASSERT_EQ(run_manual_compactions, 1);
#ifndef ROCKSDB_LITE
ASSERT_EQ("2,3,4,5,6,7,8", FilesPerLevel());
#endif // !ROCKSDB_LITE
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearCallBack(
"CompactionJob::Run():PausingManualCompaction:2");
ASSERT_OK(dbfull()->CompactRange(compact_options, nullptr, nullptr));
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
#ifndef ROCKSDB_LITE
ASSERT_EQ("0,0,0,0,0,0,2", FilesPerLevel());
#endif // !ROCKSDB_LITE
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, CancelManualCompaction1) {
CompactRangeOptions compact_options;
auto canceledPtr =
std::unique_ptr<std::atomic<bool>>(new std::atomic<bool>{true});
compact_options.canceled = canceledPtr.get();
Options options = CurrentOptions();
options.disable_auto_compactions = true;
options.num_levels = 7;
Random rnd(301);
auto generate_files = [&]() {
for (int i = 0; i < options.num_levels; i++) {
for (int j = 0; j < options.num_levels - i + 1; j++) {
for (int k = 0; k < 1000; k++) {
ASSERT_OK(Put(Key(k + j * 1000), rnd.RandomString(50)));
}
ASSERT_OK(Flush());
}
for (int l = 1; l < options.num_levels - i; l++) {
MoveFilesToLevel(l);
}
}
};
DestroyAndReopen(options);
generate_files();
#ifndef ROCKSDB_LITE
ASSERT_EQ("2,3,4,5,6,7,8", FilesPerLevel());
#endif // !ROCKSDB_LITE
int run_manual_compactions = 0;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"CompactionJob::Run():PausingManualCompaction:1",
[&](void* /*arg*/) { run_manual_compactions++; });
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
// Setup a callback to disable compactions after a couple of levels are
// compacted
int compactions_run = 0;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"DBImpl::RunManualCompaction()::1",
[&](void* /*arg*/) { ++compactions_run; });
ASSERT_TRUE(dbfull()
->CompactRange(compact_options, nullptr, nullptr)
.IsManualCompactionPaused());
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
// Since compactions are disabled, we shouldn't start compacting.
// E.g. we should call the compaction function exactly one time.
ASSERT_EQ(compactions_run, 0);
ASSERT_EQ(run_manual_compactions, 0);
#ifndef ROCKSDB_LITE
ASSERT_EQ("2,3,4,5,6,7,8", FilesPerLevel());
#endif // !ROCKSDB_LITE
compactions_run = 0;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearCallBack(
"DBImpl::RunManualCompaction()::1");
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"DBImpl::RunManualCompaction()::1", [&](void* /*arg*/) {
++compactions_run;
// After 3 compactions disable
if (compactions_run == 3) {
compact_options.canceled->store(true, std::memory_order_release);
}
});
compact_options.canceled->store(false, std::memory_order_release);
ASSERT_TRUE(dbfull()
->CompactRange(compact_options, nullptr, nullptr)
.IsManualCompactionPaused());
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
ASSERT_EQ(compactions_run, 3);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearCallBack(
"DBImpl::RunManualCompaction()::1");
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearCallBack(
"CompactionJob::Run():PausingManualCompaction:1");
// Compactions should work again if we re-enable them..
compact_options.canceled->store(false, std::memory_order_relaxed);
ASSERT_OK(dbfull()->CompactRange(compact_options, nullptr, nullptr));
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
#ifndef ROCKSDB_LITE
ASSERT_EQ("0,0,0,0,0,0,2", FilesPerLevel());
#endif // !ROCKSDB_LITE
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, CancelManualCompaction2) {
CompactRangeOptions compact_options;
auto canceledPtr =
std::unique_ptr<std::atomic<bool>>(new std::atomic<bool>{true});
compact_options.canceled = canceledPtr.get();
compact_options.max_subcompactions = 1;
Options options = CurrentOptions();
options.disable_auto_compactions = true;
options.num_levels = 7;
Random rnd(301);
auto generate_files = [&]() {
for (int i = 0; i < options.num_levels; i++) {
for (int j = 0; j < options.num_levels - i + 1; j++) {
for (int k = 0; k < 1000; k++) {
ASSERT_OK(Put(Key(k + j * 1000), rnd.RandomString(50)));
}
ASSERT_OK(Flush());
}
for (int l = 1; l < options.num_levels - i; l++) {
MoveFilesToLevel(l);
}
}
};
DestroyAndReopen(options);
generate_files();
#ifndef ROCKSDB_LITE
ASSERT_EQ("2,3,4,5,6,7,8", FilesPerLevel());
#endif // !ROCKSDB_LITE
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
int compactions_run = 0;
std::atomic<int> kv_compactions{0};
int compactions_stopped_at = 0;
int kv_compactions_stopped_at = 0;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"DBImpl::RunManualCompaction()::1", [&](void* /*arg*/) {
++compactions_run;
// After 3 compactions disable
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"CompactionIterator:ProcessKV", [&](void* /*arg*/) {
int kv_compactions_run =
kv_compactions.fetch_add(1, std::memory_order_release);
if (kv_compactions_run == 5) {
compact_options.canceled->store(true, std::memory_order_release);
kv_compactions_stopped_at = kv_compactions_run;
compactions_stopped_at = compactions_run;
}
});
compact_options.canceled->store(false, std::memory_order_release);
ASSERT_TRUE(dbfull()
->CompactRange(compact_options, nullptr, nullptr)
.IsManualCompactionPaused());
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
// NOTE: as we set compact_options.max_subcompacitons = 1, and store true to
// the canceled variable from the single compacting thread (via callback),
// this value is deterministically kv_compactions_stopped_at + 1.
ASSERT_EQ(kv_compactions, kv_compactions_stopped_at + 1);
ASSERT_EQ(compactions_run, compactions_stopped_at);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearCallBack(
"CompactionIterator::ProcessKV");
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearCallBack(
"DBImpl::RunManualCompaction()::1");
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearCallBack(
"CompactionJob::Run():PausingManualCompaction:1");
// Compactions should work again if we re-enable them..
compact_options.canceled->store(false, std::memory_order_relaxed);
ASSERT_OK(dbfull()->CompactRange(compact_options, nullptr, nullptr));
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
#ifndef ROCKSDB_LITE
ASSERT_EQ("0,0,0,0,0,0,2", FilesPerLevel());
#endif // !ROCKSDB_LITE
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
class CancelCompactionListener : public EventListener {
public:
CancelCompactionListener()
: num_compaction_started_(0), num_compaction_ended_(0) {}
void OnCompactionBegin(DB* /*db*/, const CompactionJobInfo& ci) override {
ASSERT_EQ(ci.cf_name, "default");
ASSERT_EQ(ci.base_input_level, 0);
num_compaction_started_++;
}
void OnCompactionCompleted(DB* /*db*/, const CompactionJobInfo& ci) override {
ASSERT_EQ(ci.cf_name, "default");
ASSERT_EQ(ci.base_input_level, 0);
ASSERT_EQ(ci.status.code(), code_);
ASSERT_EQ(ci.status.subcode(), subcode_);
num_compaction_ended_++;
}
std::atomic<size_t> num_compaction_started_;
std::atomic<size_t> num_compaction_ended_;
Status::Code code_;
Status::SubCode subcode_;
};
TEST_F(DBTest2, CancelManualCompactionWithListener) {
CompactRangeOptions compact_options;
auto canceledPtr =
std::unique_ptr<std::atomic<bool>>(new std::atomic<bool>{true});
compact_options.canceled = canceledPtr.get();
compact_options.max_subcompactions = 1;
Options options = CurrentOptions();
options.disable_auto_compactions = true;
CancelCompactionListener* listener = new CancelCompactionListener();
options.listeners.emplace_back(listener);
DestroyAndReopen(options);
Random rnd(301);
for (int i = 0; i < 10; i++) {
for (int j = 0; j < 10; j++) {
ASSERT_OK(Put(Key(i + j * 10), rnd.RandomString(50)));
}
ASSERT_OK(Flush());
}
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"CompactionIterator:ProcessKV", [&](void* /*arg*/) {
compact_options.canceled->store(true, std::memory_order_release);
});
int running_compaction = 0;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"CompactionJob::FinishCompactionOutputFile1",
[&](void* /*arg*/) { running_compaction++; });
// Case I: 1 Notify begin compaction, 2 Set *canceled as true to disable
// manual compaction in the callback function, 3 Compaction not run,
// 4 Notify compaction end.
listener->code_ = Status::kIncomplete;
listener->subcode_ = Status::SubCode::kManualCompactionPaused;
compact_options.canceled->store(false, std::memory_order_release);
ASSERT_TRUE(dbfull()
->CompactRange(compact_options, nullptr, nullptr)
.IsManualCompactionPaused());
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
ASSERT_GT(listener->num_compaction_started_, 0);
ASSERT_EQ(listener->num_compaction_started_, listener->num_compaction_ended_);
ASSERT_EQ(running_compaction, 0);
listener->num_compaction_started_ = 0;
listener->num_compaction_ended_ = 0;
// Case II: 1 Set *canceled as true in the callback function to disable manual
// compaction, 2 Notify begin compaction (return without notifying), 3 Notify
// compaction end (return without notifying).
ASSERT_TRUE(dbfull()
->CompactRange(compact_options, nullptr, nullptr)
.IsManualCompactionPaused());
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
ASSERT_EQ(listener->num_compaction_started_, 0);
ASSERT_EQ(listener->num_compaction_started_, listener->num_compaction_ended_);
ASSERT_EQ(running_compaction, 0);
// Case III: 1 Notify begin compaction, 2 Compaction in between
// 3. Set *canceled as true in the callback function to disable manual
// compaction, 4 Notify compaction end.
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearCallBack(
"CompactionIterator:ProcessKV");
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"CompactionJob::Run:BeforeVerify", [&](void* /*arg*/) {
compact_options.canceled->store(true, std::memory_order_release);
});
listener->code_ = Status::kOk;
listener->subcode_ = Status::SubCode::kNone;
compact_options.canceled->store(false, std::memory_order_release);
ASSERT_OK(dbfull()->CompactRange(compact_options, nullptr, nullptr));
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
ASSERT_GT(listener->num_compaction_started_, 0);
ASSERT_EQ(listener->num_compaction_started_, listener->num_compaction_ended_);
// Compaction job will succeed.
ASSERT_GT(running_compaction, 0);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearAllCallBacks();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, CompactionOnBottomPriorityWithListener) {
int num_levels = 3;
const int kNumFilesTrigger = 4;
Options options = CurrentOptions();
env_->SetBackgroundThreads(0, Env::Priority::HIGH);
env_->SetBackgroundThreads(0, Env::Priority::LOW);
env_->SetBackgroundThreads(1, Env::Priority::BOTTOM);
options.env = env_;
options.compaction_style = kCompactionStyleUniversal;
options.num_levels = num_levels;
options.write_buffer_size = 100 << 10; // 100KB
options.target_file_size_base = 32 << 10; // 32KB
options.level0_file_num_compaction_trigger = kNumFilesTrigger;
// Trigger compaction if size amplification exceeds 110%
options.compaction_options_universal.max_size_amplification_percent = 110;
CancelCompactionListener* listener = new CancelCompactionListener();
options.listeners.emplace_back(listener);
DestroyAndReopen(options);
int num_bottom_thread_compaction_scheduled = 0;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"DBImpl::BackgroundCompaction:ForwardToBottomPriPool",
[&](void* /*arg*/) { num_bottom_thread_compaction_scheduled++; });
int num_compaction_jobs = 0;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"CompactionJob::Run():End",
[&](void* /*arg*/) { num_compaction_jobs++; });
listener->code_ = Status::kOk;
listener->subcode_ = Status::SubCode::kNone;
Random rnd(301);
for (int i = 0; i < 1; ++i) {
for (int num = 0; num < kNumFilesTrigger; num++) {
int key_idx = 0;
GenerateNewFile(&rnd, &key_idx, true /* no_wait */);
// use no_wait above because that one waits for flush and compaction. We
// don't want to wait for compaction because the full compaction is
// intentionally blocked while more files are flushed.
ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable());
}
}
ASSERT_OK(dbfull()->TEST_WaitForCompact());
ASSERT_GT(num_bottom_thread_compaction_scheduled, 0);
ASSERT_EQ(num_compaction_jobs, 1);
ASSERT_GT(listener->num_compaction_started_, 0);
ASSERT_EQ(listener->num_compaction_started_, listener->num_compaction_ended_);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearAllCallBacks();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, OptimizeForPointLookup) {
Options options = CurrentOptions();
Close();
options.OptimizeForPointLookup(2);
ASSERT_OK(DB::Open(options, dbname_, &db_));
ASSERT_OK(Put("foo", "v1"));
ASSERT_EQ("v1", Get("foo"));
ASSERT_OK(Flush());
ASSERT_EQ("v1", Get("foo"));
}
TEST_F(DBTest2, OptimizeForSmallDB) {
Options options = CurrentOptions();
Close();
options.OptimizeForSmallDb();
// Find the cache object
ASSERT_TRUE(options.table_factory->IsInstanceOf(
TableFactory::kBlockBasedTableName()));
auto table_options =
options.table_factory->GetOptions<BlockBasedTableOptions>();
ASSERT_TRUE(table_options != nullptr);
std::shared_ptr<Cache> cache = table_options->block_cache;
ASSERT_EQ(0, cache->GetUsage());
ASSERT_OK(DB::Open(options, dbname_, &db_));
ASSERT_OK(Put("foo", "v1"));
// memtable size is costed to the block cache
ASSERT_NE(0, cache->GetUsage());
ASSERT_EQ("v1", Get("foo"));
ASSERT_OK(Flush());
size_t prev_size = cache->GetUsage();
// Remember block cache size, so that we can find that
// it is filled after Get().
// Use pinnable slice so that it can ping the block so that
// when we check the size it is not evicted.
PinnableSlice value;
ASSERT_OK(db_->Get(ReadOptions(), db_->DefaultColumnFamily(), "foo", &value));
ASSERT_GT(cache->GetUsage(), prev_size);
value.Reset();
}
#endif // ROCKSDB_LITE
TEST_F(DBTest2, IterRaceFlush1) {
ASSERT_OK(Put("foo", "v1"));
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->LoadDependency(
{{"DBImpl::NewIterator:1", "DBTest2::IterRaceFlush:1"},
{"DBTest2::IterRaceFlush:2", "DBImpl::NewIterator:2"}});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
ROCKSDB_NAMESPACE::port::Thread t1([&] {
TEST_SYNC_POINT("DBTest2::IterRaceFlush:1");
ASSERT_OK(Put("foo", "v2"));
ASSERT_OK(Flush());
TEST_SYNC_POINT("DBTest2::IterRaceFlush:2");
});
// iterator is created after the first Put(), and its snapshot sequence is
// assigned after second Put(), so it must see v2.
{
std::unique_ptr<Iterator> it(db_->NewIterator(ReadOptions()));
it->Seek("foo");
ASSERT_TRUE(it->Valid());
ASSERT_OK(it->status());
ASSERT_EQ("foo", it->key().ToString());
ASSERT_EQ("v2", it->value().ToString());
}
t1.join();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, IterRaceFlush2) {
ASSERT_OK(Put("foo", "v1"));
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->LoadDependency(
{{"DBImpl::NewIterator:3", "DBTest2::IterRaceFlush2:1"},
{"DBTest2::IterRaceFlush2:2", "DBImpl::NewIterator:4"}});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
ROCKSDB_NAMESPACE::port::Thread t1([&] {
TEST_SYNC_POINT("DBTest2::IterRaceFlush2:1");
ASSERT_OK(Put("foo", "v2"));
ASSERT_OK(Flush());
TEST_SYNC_POINT("DBTest2::IterRaceFlush2:2");
});
// iterator is created after the first Put(), and its snapshot sequence is
// assigned before second Put(), thus it must see v1.
{
std::unique_ptr<Iterator> it(db_->NewIterator(ReadOptions()));
it->Seek("foo");
ASSERT_TRUE(it->Valid());
ASSERT_OK(it->status());
ASSERT_EQ("foo", it->key().ToString());
ASSERT_EQ("v1", it->value().ToString());
}
t1.join();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, IterRefreshRaceFlush) {
ASSERT_OK(Put("foo", "v1"));
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->LoadDependency(
{{"ArenaWrappedDBIter::Refresh:1", "DBTest2::IterRefreshRaceFlush:1"},
{"DBTest2::IterRefreshRaceFlush:2", "ArenaWrappedDBIter::Refresh:2"}});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
ROCKSDB_NAMESPACE::port::Thread t1([&] {
TEST_SYNC_POINT("DBTest2::IterRefreshRaceFlush:1");
ASSERT_OK(Put("foo", "v2"));
ASSERT_OK(Flush());
TEST_SYNC_POINT("DBTest2::IterRefreshRaceFlush:2");
});
// iterator is refreshed after the first Put(), and its sequence number is
// assigned after second Put(), thus it must see v2.
{
std::unique_ptr<Iterator> it(db_->NewIterator(ReadOptions()));
ASSERT_OK(it->status());
ASSERT_OK(it->Refresh());
it->Seek("foo");
ASSERT_TRUE(it->Valid());
ASSERT_OK(it->status());
ASSERT_EQ("foo", it->key().ToString());
ASSERT_EQ("v2", it->value().ToString());
}
t1.join();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, GetRaceFlush1) {
ASSERT_OK(Put("foo", "v1"));
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->LoadDependency(
{{"DBImpl::GetImpl:1", "DBTest2::GetRaceFlush:1"},
{"DBTest2::GetRaceFlush:2", "DBImpl::GetImpl:2"}});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
ROCKSDB_NAMESPACE::port::Thread t1([&] {
TEST_SYNC_POINT("DBTest2::GetRaceFlush:1");
ASSERT_OK(Put("foo", "v2"));
ASSERT_OK(Flush());
TEST_SYNC_POINT("DBTest2::GetRaceFlush:2");
});
// Get() is issued after the first Put(), so it should see either
// "v1" or "v2".
ASSERT_NE("NOT_FOUND", Get("foo"));
t1.join();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, GetRaceFlush2) {
ASSERT_OK(Put("foo", "v1"));
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->LoadDependency(
{{"DBImpl::GetImpl:3", "DBTest2::GetRaceFlush:1"},
{"DBTest2::GetRaceFlush:2", "DBImpl::GetImpl:4"}});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
port::Thread t1([&] {
TEST_SYNC_POINT("DBTest2::GetRaceFlush:1");
ASSERT_OK(Put("foo", "v2"));
ASSERT_OK(Flush());
TEST_SYNC_POINT("DBTest2::GetRaceFlush:2");
});
// Get() is issued after the first Put(), so it should see either
// "v1" or "v2".
ASSERT_NE("NOT_FOUND", Get("foo"));
t1.join();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, DirectIO) {
if (!IsDirectIOSupported()) {
return;
}
Options options = CurrentOptions();
options.use_direct_reads = options.use_direct_io_for_flush_and_compaction =
true;
options.allow_mmap_reads = options.allow_mmap_writes = false;
DestroyAndReopen(options);
ASSERT_OK(Put(Key(0), "a"));
ASSERT_OK(Put(Key(5), "a"));
ASSERT_OK(Flush());
ASSERT_OK(Put(Key(10), "a"));
ASSERT_OK(Put(Key(15), "a"));
ASSERT_OK(Flush());
ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
Reopen(options);
}
TEST_F(DBTest2, MemtableOnlyIterator) {
Options options = CurrentOptions();
CreateAndReopenWithCF({"pikachu"}, options);
ASSERT_OK(Put(1, "foo", "first"));
ASSERT_OK(Put(1, "bar", "second"));
ReadOptions ropt;
ropt.read_tier = kMemtableTier;
std::string value;
Iterator* it = nullptr;
// Before flushing
// point lookups
ASSERT_OK(db_->Get(ropt, handles_[1], "foo", &value));
ASSERT_EQ("first", value);
ASSERT_OK(db_->Get(ropt, handles_[1], "bar", &value));
ASSERT_EQ("second", value);
// Memtable-only iterator (read_tier=kMemtableTier); data not flushed yet.
it = db_->NewIterator(ropt, handles_[1]);
int count = 0;
for (it->SeekToFirst(); it->Valid(); it->Next()) {
ASSERT_TRUE(it->Valid());
count++;
}
ASSERT_TRUE(!it->Valid());
ASSERT_EQ(2, count);
delete it;
Flush(1);
// After flushing
// point lookups
ASSERT_OK(db_->Get(ropt, handles_[1], "foo", &value));
ASSERT_EQ("first", value);
ASSERT_OK(db_->Get(ropt, handles_[1], "bar", &value));
ASSERT_EQ("second", value);
// nothing should be returned using memtable-only iterator after flushing.
it = db_->NewIterator(ropt, handles_[1]);
ASSERT_OK(it->status());
count = 0;
for (it->SeekToFirst(); it->Valid(); it->Next()) {
ASSERT_TRUE(it->Valid());
count++;
}
ASSERT_TRUE(!it->Valid());
ASSERT_EQ(0, count);
ASSERT_OK(it->status());
delete it;
// Add a key to memtable
ASSERT_OK(Put(1, "foobar", "third"));
it = db_->NewIterator(ropt, handles_[1]);
ASSERT_OK(it->status());
count = 0;
for (it->SeekToFirst(); it->Valid(); it->Next()) {
ASSERT_TRUE(it->Valid());
ASSERT_EQ("foobar", it->key().ToString());
ASSERT_EQ("third", it->value().ToString());
count++;
}
ASSERT_TRUE(!it->Valid());
ASSERT_EQ(1, count);
ASSERT_OK(it->status());
delete it;
}
TEST_F(DBTest2, LowPriWrite) {
Options options = CurrentOptions();
// Compaction pressure should trigger since 6 files
options.level0_file_num_compaction_trigger = 4;
options.level0_slowdown_writes_trigger = 12;
options.level0_stop_writes_trigger = 30;
options.delayed_write_rate = 8 * 1024 * 1024;
Reopen(options);
std::atomic<int> rate_limit_count(0);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"GenericRateLimiter::Request:1", [&](void* arg) {
rate_limit_count.fetch_add(1);
int64_t* rate_bytes_per_sec = static_cast<int64_t*>(arg);
ASSERT_EQ(1024 * 1024, *rate_bytes_per_sec);
});
// Block compaction
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->LoadDependency({
{"DBTest.LowPriWrite:0", "DBImpl::BGWorkCompaction"},
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
WriteOptions wo;
for (int i = 0; i < 6; i++) {
wo.low_pri = false;
ASSERT_OK(Put("", "", wo));
wo.low_pri = true;
ASSERT_OK(Put("", "", wo));
ASSERT_OK(Flush());
}
ASSERT_EQ(0, rate_limit_count.load());
wo.low_pri = true;
ASSERT_OK(Put("", "", wo));
ASSERT_EQ(1, rate_limit_count.load());
wo.low_pri = false;
ASSERT_OK(Put("", "", wo));
ASSERT_EQ(1, rate_limit_count.load());
TEST_SYNC_POINT("DBTest.LowPriWrite:0");
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
ASSERT_OK(dbfull()->TEST_WaitForCompact());
wo.low_pri = true;
ASSERT_OK(Put("", "", wo));
ASSERT_EQ(1, rate_limit_count.load());
wo.low_pri = false;
ASSERT_OK(Put("", "", wo));
ASSERT_EQ(1, rate_limit_count.load());
}
#ifndef ROCKSDB_LITE
TEST_F(DBTest2, RateLimitedCompactionReads) {
// compaction input has 512KB data
const int kNumKeysPerFile = 128;
const int kBytesPerKey = 1024;
const int kNumL0Files = 4;
Add rate limiter priority to ReadOptions (#9424) Summary: Users can set the priority for file reads associated with their operation by setting `ReadOptions::rate_limiter_priority` to something other than `Env::IO_TOTAL`. Rate limiting `VerifyChecksum()` and `VerifyFileChecksums()` is the motivation for this PR, so it also includes benchmarks and minor bug fixes to get that working. `RandomAccessFileReader::Read()` already had support for rate limiting compaction reads. I changed that rate limiting to be non-specific to compaction, but rather performed according to the passed in `Env::IOPriority`. Now the compaction read rate limiting is supported by setting `rate_limiter_priority = Env::IO_LOW` on its `ReadOptions`. There is no default value for the new `Env::IOPriority` parameter to `RandomAccessFileReader::Read()`. That means this PR goes through all callers (in some cases multiple layers up the call stack) to find a `ReadOptions` to provide the priority. There are TODOs for cases I believe it would be good to let user control the priority some day (e.g., file footer reads), and no TODO in cases I believe it doesn't matter (e.g., trace file reads). The API doc only lists the missing cases where a file read associated with a provided `ReadOptions` cannot be rate limited. For cases like file ingestion checksum calculation, there is no API to provide `ReadOptions` or `Env::IOPriority`, so I didn't count that as missing. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9424 Test Plan: - new unit tests - new benchmarks on ~50MB database with 1MB/s read rate limit and 100ms refill interval; verified with strace reads are chunked (at 0.1MB per chunk) and spaced roughly 100ms apart. - setup command: `./db_bench -benchmarks=fillrandom,compact -db=/tmp/testdb -target_file_size_base=1048576 -disable_auto_compactions=true -file_checksum=true` - benchmarks command: `strace -ttfe pread64 ./db_bench -benchmarks=verifychecksum,verifyfilechecksums -use_existing_db=true -db=/tmp/testdb -rate_limiter_bytes_per_sec=1048576 -rate_limit_bg_reads=1 -rate_limit_user_ops=true -file_checksum=true` - crash test using IO_USER priority on non-validation reads with https://github.com/facebook/rocksdb/issues/9567 reverted: `python3 tools/db_crashtest.py blackbox --max_key=1000000 --write_buffer_size=524288 --target_file_size_base=524288 --level_compaction_dynamic_level_bytes=true --duration=3600 --rate_limit_bg_reads=true --rate_limit_user_ops=true --rate_limiter_bytes_per_sec=10485760 --interval=10` Reviewed By: hx235 Differential Revision: D33747386 Pulled By: ajkr fbshipit-source-id: a2d985e97912fba8c54763798e04f006ccc56e0c
3 years ago
for (int compaction_readahead_size : {0, 32 << 10}) {
for (auto use_direct_io : {false, true}) {
if (use_direct_io && !IsDirectIOSupported()) {
continue;
}
Options options = CurrentOptions();
options.compaction_readahead_size = compaction_readahead_size;
options.compression = kNoCompression;
options.level0_file_num_compaction_trigger = kNumL0Files;
options.memtable_factory.reset(
test::NewSpecialSkipListFactory(kNumKeysPerFile));
// takes roughly one second, split into 100 x 10ms intervals. Each
// interval permits 5.12KB, which is smaller than the block size, so this
// test exercises the code for chunking reads.
options.rate_limiter.reset(NewGenericRateLimiter(
static_cast<int64_t>(kNumL0Files * kNumKeysPerFile *
kBytesPerKey) /* rate_bytes_per_sec */,
10 * 1000 /* refill_period_us */, 10 /* fairness */,
RateLimiter::Mode::kReadsOnly));
options.use_direct_reads =
options.use_direct_io_for_flush_and_compaction = use_direct_io;
BlockBasedTableOptions bbto;
bbto.block_size = 16384;
bbto.no_block_cache = true;
options.table_factory.reset(NewBlockBasedTableFactory(bbto));
DestroyAndReopen(options);
Add rate limiter priority to ReadOptions (#9424) Summary: Users can set the priority for file reads associated with their operation by setting `ReadOptions::rate_limiter_priority` to something other than `Env::IO_TOTAL`. Rate limiting `VerifyChecksum()` and `VerifyFileChecksums()` is the motivation for this PR, so it also includes benchmarks and minor bug fixes to get that working. `RandomAccessFileReader::Read()` already had support for rate limiting compaction reads. I changed that rate limiting to be non-specific to compaction, but rather performed according to the passed in `Env::IOPriority`. Now the compaction read rate limiting is supported by setting `rate_limiter_priority = Env::IO_LOW` on its `ReadOptions`. There is no default value for the new `Env::IOPriority` parameter to `RandomAccessFileReader::Read()`. That means this PR goes through all callers (in some cases multiple layers up the call stack) to find a `ReadOptions` to provide the priority. There are TODOs for cases I believe it would be good to let user control the priority some day (e.g., file footer reads), and no TODO in cases I believe it doesn't matter (e.g., trace file reads). The API doc only lists the missing cases where a file read associated with a provided `ReadOptions` cannot be rate limited. For cases like file ingestion checksum calculation, there is no API to provide `ReadOptions` or `Env::IOPriority`, so I didn't count that as missing. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9424 Test Plan: - new unit tests - new benchmarks on ~50MB database with 1MB/s read rate limit and 100ms refill interval; verified with strace reads are chunked (at 0.1MB per chunk) and spaced roughly 100ms apart. - setup command: `./db_bench -benchmarks=fillrandom,compact -db=/tmp/testdb -target_file_size_base=1048576 -disable_auto_compactions=true -file_checksum=true` - benchmarks command: `strace -ttfe pread64 ./db_bench -benchmarks=verifychecksum,verifyfilechecksums -use_existing_db=true -db=/tmp/testdb -rate_limiter_bytes_per_sec=1048576 -rate_limit_bg_reads=1 -rate_limit_user_ops=true -file_checksum=true` - crash test using IO_USER priority on non-validation reads with https://github.com/facebook/rocksdb/issues/9567 reverted: `python3 tools/db_crashtest.py blackbox --max_key=1000000 --write_buffer_size=524288 --target_file_size_base=524288 --level_compaction_dynamic_level_bytes=true --duration=3600 --rate_limit_bg_reads=true --rate_limit_user_ops=true --rate_limiter_bytes_per_sec=10485760 --interval=10` Reviewed By: hx235 Differential Revision: D33747386 Pulled By: ajkr fbshipit-source-id: a2d985e97912fba8c54763798e04f006ccc56e0c
3 years ago
for (int i = 0; i < kNumL0Files; ++i) {
for (int j = 0; j <= kNumKeysPerFile; ++j) {
ASSERT_OK(Put(Key(j), DummyString(kBytesPerKey)));
}
ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable());
if (i + 1 < kNumL0Files) {
ASSERT_EQ(i + 1, NumTableFilesAtLevel(0));
}
}
Add rate limiter priority to ReadOptions (#9424) Summary: Users can set the priority for file reads associated with their operation by setting `ReadOptions::rate_limiter_priority` to something other than `Env::IO_TOTAL`. Rate limiting `VerifyChecksum()` and `VerifyFileChecksums()` is the motivation for this PR, so it also includes benchmarks and minor bug fixes to get that working. `RandomAccessFileReader::Read()` already had support for rate limiting compaction reads. I changed that rate limiting to be non-specific to compaction, but rather performed according to the passed in `Env::IOPriority`. Now the compaction read rate limiting is supported by setting `rate_limiter_priority = Env::IO_LOW` on its `ReadOptions`. There is no default value for the new `Env::IOPriority` parameter to `RandomAccessFileReader::Read()`. That means this PR goes through all callers (in some cases multiple layers up the call stack) to find a `ReadOptions` to provide the priority. There are TODOs for cases I believe it would be good to let user control the priority some day (e.g., file footer reads), and no TODO in cases I believe it doesn't matter (e.g., trace file reads). The API doc only lists the missing cases where a file read associated with a provided `ReadOptions` cannot be rate limited. For cases like file ingestion checksum calculation, there is no API to provide `ReadOptions` or `Env::IOPriority`, so I didn't count that as missing. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9424 Test Plan: - new unit tests - new benchmarks on ~50MB database with 1MB/s read rate limit and 100ms refill interval; verified with strace reads are chunked (at 0.1MB per chunk) and spaced roughly 100ms apart. - setup command: `./db_bench -benchmarks=fillrandom,compact -db=/tmp/testdb -target_file_size_base=1048576 -disable_auto_compactions=true -file_checksum=true` - benchmarks command: `strace -ttfe pread64 ./db_bench -benchmarks=verifychecksum,verifyfilechecksums -use_existing_db=true -db=/tmp/testdb -rate_limiter_bytes_per_sec=1048576 -rate_limit_bg_reads=1 -rate_limit_user_ops=true -file_checksum=true` - crash test using IO_USER priority on non-validation reads with https://github.com/facebook/rocksdb/issues/9567 reverted: `python3 tools/db_crashtest.py blackbox --max_key=1000000 --write_buffer_size=524288 --target_file_size_base=524288 --level_compaction_dynamic_level_bytes=true --duration=3600 --rate_limit_bg_reads=true --rate_limit_user_ops=true --rate_limiter_bytes_per_sec=10485760 --interval=10` Reviewed By: hx235 Differential Revision: D33747386 Pulled By: ajkr fbshipit-source-id: a2d985e97912fba8c54763798e04f006ccc56e0c
3 years ago
ASSERT_OK(dbfull()->TEST_WaitForCompact());
ASSERT_EQ(0, NumTableFilesAtLevel(0));
// should be slightly above 512KB due to non-data blocks read. Arbitrarily
// chose 1MB as the upper bound on the total bytes read.
Set Read rate limiter priority dynamically and pass it to FS (#9996) Summary: ### Context: Background compactions and flush generate large reads and writes, and can be long running, especially for universal compaction. In some cases, this can impact foreground reads and writes by users. ### Solution User, Flush, and Compaction reads share some code path. For this task, we update the rate_limiter_priority in ReadOptions for code paths (e.g. FindTable (mainly in BlockBasedTable::Open()) and various iterators), and eventually update the rate_limiter_priority in IOOptions for FSRandomAccessFile. **This PR is for the Read path.** The **Read:** dynamic priority for different state are listed as follows: | State | Normal | Delayed | Stalled | | ----- | ------ | ------- | ------- | | Flush (verification read in BuildTable()) | IO_USER | IO_USER | IO_USER | | Compaction | IO_LOW | IO_USER | IO_USER | | User | User provided | User provided | User provided | We will respect the read_options that the user provided and will not set it. The only sst read for Flush is the verification read in BuildTable(). It claims to be "regard as user read". **Details** 1. Set read_options.rate_limiter_priority dynamically: - User: Do not update the read_options. Use the read_options that the user provided. - Compaction: Update read_options in CompactionJob::ProcessKeyValueCompaction(). - Flush: Update read_options in BuildTable(). 2. Pass the rate limiter priority to FSRandomAccessFile functions: - After calling the FindTable(), read_options is passed through GetTableReader(table_cache.cc), BlockBasedTableFactory::NewTableReader(block_based_table_factory.cc), and BlockBasedTable::Open(). The Open() needs some updates for the ReadOptions variable and the updates are also needed for the called functions, including PrefetchTail(), PrepareIOOptions(), ReadFooterFromFile(), ReadMetaIndexblock(), ReadPropertiesBlock(), PrefetchIndexAndFilterBlocks(), and ReadRangeDelBlock(). - In RandomAccessFileReader, the functions to be updated include Read(), MultiRead(), ReadAsync(), and Prefetch(). - Update the downstream functions of NewIndexIterator(), NewDataBlockIterator(), and BlockBasedTableIterator(). ### Test Plans Add unit tests. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9996 Reviewed By: anand1976 Differential Revision: D36452483 Pulled By: gitbw95 fbshipit-source-id: 60978204a4f849bb9261cb78d9bc1cb56d6008cf
3 years ago
size_t rate_limited_bytes = static_cast<size_t>(
options.rate_limiter->GetTotalBytesThrough(Env::IO_TOTAL));
// The charges can exist for `IO_LOW` and `IO_USER` priorities.
size_t rate_limited_bytes_by_pri =
options.rate_limiter->GetTotalBytesThrough(Env::IO_LOW) +
options.rate_limiter->GetTotalBytesThrough(Env::IO_USER);
Add rate limiter priority to ReadOptions (#9424) Summary: Users can set the priority for file reads associated with their operation by setting `ReadOptions::rate_limiter_priority` to something other than `Env::IO_TOTAL`. Rate limiting `VerifyChecksum()` and `VerifyFileChecksums()` is the motivation for this PR, so it also includes benchmarks and minor bug fixes to get that working. `RandomAccessFileReader::Read()` already had support for rate limiting compaction reads. I changed that rate limiting to be non-specific to compaction, but rather performed according to the passed in `Env::IOPriority`. Now the compaction read rate limiting is supported by setting `rate_limiter_priority = Env::IO_LOW` on its `ReadOptions`. There is no default value for the new `Env::IOPriority` parameter to `RandomAccessFileReader::Read()`. That means this PR goes through all callers (in some cases multiple layers up the call stack) to find a `ReadOptions` to provide the priority. There are TODOs for cases I believe it would be good to let user control the priority some day (e.g., file footer reads), and no TODO in cases I believe it doesn't matter (e.g., trace file reads). The API doc only lists the missing cases where a file read associated with a provided `ReadOptions` cannot be rate limited. For cases like file ingestion checksum calculation, there is no API to provide `ReadOptions` or `Env::IOPriority`, so I didn't count that as missing. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9424 Test Plan: - new unit tests - new benchmarks on ~50MB database with 1MB/s read rate limit and 100ms refill interval; verified with strace reads are chunked (at 0.1MB per chunk) and spaced roughly 100ms apart. - setup command: `./db_bench -benchmarks=fillrandom,compact -db=/tmp/testdb -target_file_size_base=1048576 -disable_auto_compactions=true -file_checksum=true` - benchmarks command: `strace -ttfe pread64 ./db_bench -benchmarks=verifychecksum,verifyfilechecksums -use_existing_db=true -db=/tmp/testdb -rate_limiter_bytes_per_sec=1048576 -rate_limit_bg_reads=1 -rate_limit_user_ops=true -file_checksum=true` - crash test using IO_USER priority on non-validation reads with https://github.com/facebook/rocksdb/issues/9567 reverted: `python3 tools/db_crashtest.py blackbox --max_key=1000000 --write_buffer_size=524288 --target_file_size_base=524288 --level_compaction_dynamic_level_bytes=true --duration=3600 --rate_limit_bg_reads=true --rate_limit_user_ops=true --rate_limiter_bytes_per_sec=10485760 --interval=10` Reviewed By: hx235 Differential Revision: D33747386 Pulled By: ajkr fbshipit-source-id: a2d985e97912fba8c54763798e04f006ccc56e0c
3 years ago
ASSERT_EQ(rate_limited_bytes,
Set Read rate limiter priority dynamically and pass it to FS (#9996) Summary: ### Context: Background compactions and flush generate large reads and writes, and can be long running, especially for universal compaction. In some cases, this can impact foreground reads and writes by users. ### Solution User, Flush, and Compaction reads share some code path. For this task, we update the rate_limiter_priority in ReadOptions for code paths (e.g. FindTable (mainly in BlockBasedTable::Open()) and various iterators), and eventually update the rate_limiter_priority in IOOptions for FSRandomAccessFile. **This PR is for the Read path.** The **Read:** dynamic priority for different state are listed as follows: | State | Normal | Delayed | Stalled | | ----- | ------ | ------- | ------- | | Flush (verification read in BuildTable()) | IO_USER | IO_USER | IO_USER | | Compaction | IO_LOW | IO_USER | IO_USER | | User | User provided | User provided | User provided | We will respect the read_options that the user provided and will not set it. The only sst read for Flush is the verification read in BuildTable(). It claims to be "regard as user read". **Details** 1. Set read_options.rate_limiter_priority dynamically: - User: Do not update the read_options. Use the read_options that the user provided. - Compaction: Update read_options in CompactionJob::ProcessKeyValueCompaction(). - Flush: Update read_options in BuildTable(). 2. Pass the rate limiter priority to FSRandomAccessFile functions: - After calling the FindTable(), read_options is passed through GetTableReader(table_cache.cc), BlockBasedTableFactory::NewTableReader(block_based_table_factory.cc), and BlockBasedTable::Open(). The Open() needs some updates for the ReadOptions variable and the updates are also needed for the called functions, including PrefetchTail(), PrepareIOOptions(), ReadFooterFromFile(), ReadMetaIndexblock(), ReadPropertiesBlock(), PrefetchIndexAndFilterBlocks(), and ReadRangeDelBlock(). - In RandomAccessFileReader, the functions to be updated include Read(), MultiRead(), ReadAsync(), and Prefetch(). - Update the downstream functions of NewIndexIterator(), NewDataBlockIterator(), and BlockBasedTableIterator(). ### Test Plans Add unit tests. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9996 Reviewed By: anand1976 Differential Revision: D36452483 Pulled By: gitbw95 fbshipit-source-id: 60978204a4f849bb9261cb78d9bc1cb56d6008cf
3 years ago
static_cast<size_t>(rate_limited_bytes_by_pri));
Add rate limiter priority to ReadOptions (#9424) Summary: Users can set the priority for file reads associated with their operation by setting `ReadOptions::rate_limiter_priority` to something other than `Env::IO_TOTAL`. Rate limiting `VerifyChecksum()` and `VerifyFileChecksums()` is the motivation for this PR, so it also includes benchmarks and minor bug fixes to get that working. `RandomAccessFileReader::Read()` already had support for rate limiting compaction reads. I changed that rate limiting to be non-specific to compaction, but rather performed according to the passed in `Env::IOPriority`. Now the compaction read rate limiting is supported by setting `rate_limiter_priority = Env::IO_LOW` on its `ReadOptions`. There is no default value for the new `Env::IOPriority` parameter to `RandomAccessFileReader::Read()`. That means this PR goes through all callers (in some cases multiple layers up the call stack) to find a `ReadOptions` to provide the priority. There are TODOs for cases I believe it would be good to let user control the priority some day (e.g., file footer reads), and no TODO in cases I believe it doesn't matter (e.g., trace file reads). The API doc only lists the missing cases where a file read associated with a provided `ReadOptions` cannot be rate limited. For cases like file ingestion checksum calculation, there is no API to provide `ReadOptions` or `Env::IOPriority`, so I didn't count that as missing. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9424 Test Plan: - new unit tests - new benchmarks on ~50MB database with 1MB/s read rate limit and 100ms refill interval; verified with strace reads are chunked (at 0.1MB per chunk) and spaced roughly 100ms apart. - setup command: `./db_bench -benchmarks=fillrandom,compact -db=/tmp/testdb -target_file_size_base=1048576 -disable_auto_compactions=true -file_checksum=true` - benchmarks command: `strace -ttfe pread64 ./db_bench -benchmarks=verifychecksum,verifyfilechecksums -use_existing_db=true -db=/tmp/testdb -rate_limiter_bytes_per_sec=1048576 -rate_limit_bg_reads=1 -rate_limit_user_ops=true -file_checksum=true` - crash test using IO_USER priority on non-validation reads with https://github.com/facebook/rocksdb/issues/9567 reverted: `python3 tools/db_crashtest.py blackbox --max_key=1000000 --write_buffer_size=524288 --target_file_size_base=524288 --level_compaction_dynamic_level_bytes=true --duration=3600 --rate_limit_bg_reads=true --rate_limit_user_ops=true --rate_limiter_bytes_per_sec=10485760 --interval=10` Reviewed By: hx235 Differential Revision: D33747386 Pulled By: ajkr fbshipit-source-id: a2d985e97912fba8c54763798e04f006ccc56e0c
3 years ago
// Include the explicit prefetch of the footer in direct I/O case.
size_t direct_io_extra = use_direct_io ? 512 * 1024 : 0;
ASSERT_GE(
rate_limited_bytes,
static_cast<size_t>(kNumKeysPerFile * kBytesPerKey * kNumL0Files));
ASSERT_LT(
rate_limited_bytes,
static_cast<size_t>(2 * kNumKeysPerFile * kBytesPerKey * kNumL0Files +
direct_io_extra));
Iterator* iter = db_->NewIterator(ReadOptions());
ASSERT_OK(iter->status());
for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
ASSERT_EQ(iter->value().ToString(), DummyString(kBytesPerKey));
}
Add rate limiter priority to ReadOptions (#9424) Summary: Users can set the priority for file reads associated with their operation by setting `ReadOptions::rate_limiter_priority` to something other than `Env::IO_TOTAL`. Rate limiting `VerifyChecksum()` and `VerifyFileChecksums()` is the motivation for this PR, so it also includes benchmarks and minor bug fixes to get that working. `RandomAccessFileReader::Read()` already had support for rate limiting compaction reads. I changed that rate limiting to be non-specific to compaction, but rather performed according to the passed in `Env::IOPriority`. Now the compaction read rate limiting is supported by setting `rate_limiter_priority = Env::IO_LOW` on its `ReadOptions`. There is no default value for the new `Env::IOPriority` parameter to `RandomAccessFileReader::Read()`. That means this PR goes through all callers (in some cases multiple layers up the call stack) to find a `ReadOptions` to provide the priority. There are TODOs for cases I believe it would be good to let user control the priority some day (e.g., file footer reads), and no TODO in cases I believe it doesn't matter (e.g., trace file reads). The API doc only lists the missing cases where a file read associated with a provided `ReadOptions` cannot be rate limited. For cases like file ingestion checksum calculation, there is no API to provide `ReadOptions` or `Env::IOPriority`, so I didn't count that as missing. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9424 Test Plan: - new unit tests - new benchmarks on ~50MB database with 1MB/s read rate limit and 100ms refill interval; verified with strace reads are chunked (at 0.1MB per chunk) and spaced roughly 100ms apart. - setup command: `./db_bench -benchmarks=fillrandom,compact -db=/tmp/testdb -target_file_size_base=1048576 -disable_auto_compactions=true -file_checksum=true` - benchmarks command: `strace -ttfe pread64 ./db_bench -benchmarks=verifychecksum,verifyfilechecksums -use_existing_db=true -db=/tmp/testdb -rate_limiter_bytes_per_sec=1048576 -rate_limit_bg_reads=1 -rate_limit_user_ops=true -file_checksum=true` - crash test using IO_USER priority on non-validation reads with https://github.com/facebook/rocksdb/issues/9567 reverted: `python3 tools/db_crashtest.py blackbox --max_key=1000000 --write_buffer_size=524288 --target_file_size_base=524288 --level_compaction_dynamic_level_bytes=true --duration=3600 --rate_limit_bg_reads=true --rate_limit_user_ops=true --rate_limiter_bytes_per_sec=10485760 --interval=10` Reviewed By: hx235 Differential Revision: D33747386 Pulled By: ajkr fbshipit-source-id: a2d985e97912fba8c54763798e04f006ccc56e0c
3 years ago
delete iter;
// bytes read for user iterator shouldn't count against the rate limit.
Set Read rate limiter priority dynamically and pass it to FS (#9996) Summary: ### Context: Background compactions and flush generate large reads and writes, and can be long running, especially for universal compaction. In some cases, this can impact foreground reads and writes by users. ### Solution User, Flush, and Compaction reads share some code path. For this task, we update the rate_limiter_priority in ReadOptions for code paths (e.g. FindTable (mainly in BlockBasedTable::Open()) and various iterators), and eventually update the rate_limiter_priority in IOOptions for FSRandomAccessFile. **This PR is for the Read path.** The **Read:** dynamic priority for different state are listed as follows: | State | Normal | Delayed | Stalled | | ----- | ------ | ------- | ------- | | Flush (verification read in BuildTable()) | IO_USER | IO_USER | IO_USER | | Compaction | IO_LOW | IO_USER | IO_USER | | User | User provided | User provided | User provided | We will respect the read_options that the user provided and will not set it. The only sst read for Flush is the verification read in BuildTable(). It claims to be "regard as user read". **Details** 1. Set read_options.rate_limiter_priority dynamically: - User: Do not update the read_options. Use the read_options that the user provided. - Compaction: Update read_options in CompactionJob::ProcessKeyValueCompaction(). - Flush: Update read_options in BuildTable(). 2. Pass the rate limiter priority to FSRandomAccessFile functions: - After calling the FindTable(), read_options is passed through GetTableReader(table_cache.cc), BlockBasedTableFactory::NewTableReader(block_based_table_factory.cc), and BlockBasedTable::Open(). The Open() needs some updates for the ReadOptions variable and the updates are also needed for the called functions, including PrefetchTail(), PrepareIOOptions(), ReadFooterFromFile(), ReadMetaIndexblock(), ReadPropertiesBlock(), PrefetchIndexAndFilterBlocks(), and ReadRangeDelBlock(). - In RandomAccessFileReader, the functions to be updated include Read(), MultiRead(), ReadAsync(), and Prefetch(). - Update the downstream functions of NewIndexIterator(), NewDataBlockIterator(), and BlockBasedTableIterator(). ### Test Plans Add unit tests. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9996 Reviewed By: anand1976 Differential Revision: D36452483 Pulled By: gitbw95 fbshipit-source-id: 60978204a4f849bb9261cb78d9bc1cb56d6008cf
3 years ago
rate_limited_bytes_by_pri =
options.rate_limiter->GetTotalBytesThrough(Env::IO_LOW) +
options.rate_limiter->GetTotalBytesThrough(Env::IO_USER);
Add rate limiter priority to ReadOptions (#9424) Summary: Users can set the priority for file reads associated with their operation by setting `ReadOptions::rate_limiter_priority` to something other than `Env::IO_TOTAL`. Rate limiting `VerifyChecksum()` and `VerifyFileChecksums()` is the motivation for this PR, so it also includes benchmarks and minor bug fixes to get that working. `RandomAccessFileReader::Read()` already had support for rate limiting compaction reads. I changed that rate limiting to be non-specific to compaction, but rather performed according to the passed in `Env::IOPriority`. Now the compaction read rate limiting is supported by setting `rate_limiter_priority = Env::IO_LOW` on its `ReadOptions`. There is no default value for the new `Env::IOPriority` parameter to `RandomAccessFileReader::Read()`. That means this PR goes through all callers (in some cases multiple layers up the call stack) to find a `ReadOptions` to provide the priority. There are TODOs for cases I believe it would be good to let user control the priority some day (e.g., file footer reads), and no TODO in cases I believe it doesn't matter (e.g., trace file reads). The API doc only lists the missing cases where a file read associated with a provided `ReadOptions` cannot be rate limited. For cases like file ingestion checksum calculation, there is no API to provide `ReadOptions` or `Env::IOPriority`, so I didn't count that as missing. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9424 Test Plan: - new unit tests - new benchmarks on ~50MB database with 1MB/s read rate limit and 100ms refill interval; verified with strace reads are chunked (at 0.1MB per chunk) and spaced roughly 100ms apart. - setup command: `./db_bench -benchmarks=fillrandom,compact -db=/tmp/testdb -target_file_size_base=1048576 -disable_auto_compactions=true -file_checksum=true` - benchmarks command: `strace -ttfe pread64 ./db_bench -benchmarks=verifychecksum,verifyfilechecksums -use_existing_db=true -db=/tmp/testdb -rate_limiter_bytes_per_sec=1048576 -rate_limit_bg_reads=1 -rate_limit_user_ops=true -file_checksum=true` - crash test using IO_USER priority on non-validation reads with https://github.com/facebook/rocksdb/issues/9567 reverted: `python3 tools/db_crashtest.py blackbox --max_key=1000000 --write_buffer_size=524288 --target_file_size_base=524288 --level_compaction_dynamic_level_bytes=true --duration=3600 --rate_limit_bg_reads=true --rate_limit_user_ops=true --rate_limiter_bytes_per_sec=10485760 --interval=10` Reviewed By: hx235 Differential Revision: D33747386 Pulled By: ajkr fbshipit-source-id: a2d985e97912fba8c54763798e04f006ccc56e0c
3 years ago
ASSERT_EQ(rate_limited_bytes,
Set Read rate limiter priority dynamically and pass it to FS (#9996) Summary: ### Context: Background compactions and flush generate large reads and writes, and can be long running, especially for universal compaction. In some cases, this can impact foreground reads and writes by users. ### Solution User, Flush, and Compaction reads share some code path. For this task, we update the rate_limiter_priority in ReadOptions for code paths (e.g. FindTable (mainly in BlockBasedTable::Open()) and various iterators), and eventually update the rate_limiter_priority in IOOptions for FSRandomAccessFile. **This PR is for the Read path.** The **Read:** dynamic priority for different state are listed as follows: | State | Normal | Delayed | Stalled | | ----- | ------ | ------- | ------- | | Flush (verification read in BuildTable()) | IO_USER | IO_USER | IO_USER | | Compaction | IO_LOW | IO_USER | IO_USER | | User | User provided | User provided | User provided | We will respect the read_options that the user provided and will not set it. The only sst read for Flush is the verification read in BuildTable(). It claims to be "regard as user read". **Details** 1. Set read_options.rate_limiter_priority dynamically: - User: Do not update the read_options. Use the read_options that the user provided. - Compaction: Update read_options in CompactionJob::ProcessKeyValueCompaction(). - Flush: Update read_options in BuildTable(). 2. Pass the rate limiter priority to FSRandomAccessFile functions: - After calling the FindTable(), read_options is passed through GetTableReader(table_cache.cc), BlockBasedTableFactory::NewTableReader(block_based_table_factory.cc), and BlockBasedTable::Open(). The Open() needs some updates for the ReadOptions variable and the updates are also needed for the called functions, including PrefetchTail(), PrepareIOOptions(), ReadFooterFromFile(), ReadMetaIndexblock(), ReadPropertiesBlock(), PrefetchIndexAndFilterBlocks(), and ReadRangeDelBlock(). - In RandomAccessFileReader, the functions to be updated include Read(), MultiRead(), ReadAsync(), and Prefetch(). - Update the downstream functions of NewIndexIterator(), NewDataBlockIterator(), and BlockBasedTableIterator(). ### Test Plans Add unit tests. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9996 Reviewed By: anand1976 Differential Revision: D36452483 Pulled By: gitbw95 fbshipit-source-id: 60978204a4f849bb9261cb78d9bc1cb56d6008cf
3 years ago
static_cast<size_t>(rate_limited_bytes_by_pri));
}
}
}
#endif // ROCKSDB_LITE
// Make sure DB can be reopen with reduced number of levels, given no file
// is on levels higher than the new num_levels.
TEST_F(DBTest2, ReduceLevel) {
Options options;
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
options.env = env_;
options.disable_auto_compactions = true;
options.num_levels = 7;
Reopen(options);
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Flush());
MoveFilesToLevel(6);
#ifndef ROCKSDB_LITE
ASSERT_EQ("0,0,0,0,0,0,1", FilesPerLevel());
#endif // !ROCKSDB_LITE
CompactRangeOptions compact_options;
compact_options.change_level = true;
compact_options.target_level = 1;
ASSERT_OK(dbfull()->CompactRange(compact_options, nullptr, nullptr));
#ifndef ROCKSDB_LITE
ASSERT_EQ("0,1", FilesPerLevel());
#endif // !ROCKSDB_LITE
options.num_levels = 3;
Reopen(options);
#ifndef ROCKSDB_LITE
ASSERT_EQ("0,1", FilesPerLevel());
#endif // !ROCKSDB_LITE
}
// Test that ReadCallback is actually used in both memtbale and sst tables
TEST_F(DBTest2, ReadCallbackTest) {
Options options;
options.disable_auto_compactions = true;
options.num_levels = 7;
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
options.env = env_;
Reopen(options);
std::vector<const Snapshot*> snapshots;
// Try to create a db with multiple layers and a memtable
const std::string key = "foo";
const std::string value = "bar";
// This test assumes that the seq start with 1 and increased by 1 after each
// write batch of size 1. If that behavior changes, the test needs to be
// updated as well.
// TODO(myabandeh): update this test to use the seq number that is returned by
// the DB instead of assuming what seq the DB used.
int i = 1;
for (; i < 10; i++) {
ASSERT_OK(Put(key, value + std::to_string(i)));
// Take a snapshot to avoid the value being removed during compaction
auto snapshot = dbfull()->GetSnapshot();
snapshots.push_back(snapshot);
}
ASSERT_OK(Flush());
for (; i < 20; i++) {
ASSERT_OK(Put(key, value + std::to_string(i)));
// Take a snapshot to avoid the value being removed during compaction
auto snapshot = dbfull()->GetSnapshot();
snapshots.push_back(snapshot);
}
ASSERT_OK(Flush());
MoveFilesToLevel(6);
#ifndef ROCKSDB_LITE
ASSERT_EQ("0,0,0,0,0,0,2", FilesPerLevel());
#endif // !ROCKSDB_LITE
for (; i < 30; i++) {
ASSERT_OK(Put(key, value + std::to_string(i)));
auto snapshot = dbfull()->GetSnapshot();
snapshots.push_back(snapshot);
}
ASSERT_OK(Flush());
#ifndef ROCKSDB_LITE
ASSERT_EQ("1,0,0,0,0,0,2", FilesPerLevel());
#endif // !ROCKSDB_LITE
// And also add some values to the memtable
for (; i < 40; i++) {
ASSERT_OK(Put(key, value + std::to_string(i)));
auto snapshot = dbfull()->GetSnapshot();
snapshots.push_back(snapshot);
}
class TestReadCallback : public ReadCallback {
public:
WriteUnPrepared: less virtual in iterator callback (#5049) Summary: WriteUnPrepared adds a virtual function, MaxUnpreparedSequenceNumber, to ReadCallback, which returns 0 unless WriteUnPrepared is enabled and the transaction has uncommitted data written to the DB. Together with snapshot sequence number, this determines the last sequence that is visible to reads. The patch clarifies the guarantees of the GetIterator API in WriteUnPrepared transactions and make use of that to statically initialize the read callback and thus avoid the virtual call. Furthermore it increases the minimum value for min_uncommitted from 0 to 1 as seq 0 is used only for last level keys that are committed in all snapshots. The following benchmark shows +0.26% higher throughput in seekrandom benchmark. Benchmark: ./db_bench --benchmarks=fillrandom --use_existing_db=0 --num=1000000 --db=/dev/shm/dbbench ./db_bench --benchmarks=seekrandom[X10] --use_existing_db=1 --db=/dev/shm/dbbench --num=1000000 --duration=60 --seek_nexts=100 seekrandom [AVG 10 runs] : 20355 ops/sec; 225.2 MB/sec seekrandom [MEDIAN 10 runs] : 20425 ops/sec; 225.9 MB/sec ./db_bench_lessvirtual3 --benchmarks=seekrandom[X10] --use_existing_db=1 --db=/dev/shm/dbbench --num=1000000 --duration=60 --seek_nexts=100 seekrandom [AVG 10 runs] : 20409 ops/sec; 225.8 MB/sec seekrandom [MEDIAN 10 runs] : 20487 ops/sec; 226.6 MB/sec Pull Request resolved: https://github.com/facebook/rocksdb/pull/5049 Differential Revision: D14366459 Pulled By: maysamyabandeh fbshipit-source-id: ebaff8908332a5ae9af7defeadabcb624be660ef
6 years ago
explicit TestReadCallback(SequenceNumber snapshot)
: ReadCallback(snapshot), snapshot_(snapshot) {}
bool IsVisibleFullCheck(SequenceNumber seq) override {
return seq <= snapshot_;
}
private:
SequenceNumber snapshot_;
};
for (int seq = 1; seq < i; seq++) {
PinnableSlice pinnable_val;
ReadOptions roptions;
TestReadCallback callback(seq);
bool dont_care = true;
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
DBImpl::GetImplOptions get_impl_options;
get_impl_options.column_family = dbfull()->DefaultColumnFamily();
get_impl_options.value = &pinnable_val;
get_impl_options.value_found = &dont_care;
get_impl_options.callback = &callback;
Status s = dbfull()->GetImpl(roptions, key, get_impl_options);
ASSERT_TRUE(s.ok());
// Assuming that after each Put the DB increased seq by one, the value and
// seq number must be equal since we also inc value by 1 after each Put.
ASSERT_EQ(value + std::to_string(seq), pinnable_val.ToString());
}
for (auto snapshot : snapshots) {
dbfull()->ReleaseSnapshot(snapshot);
}
}
#ifndef ROCKSDB_LITE
TEST_F(DBTest2, LiveFilesOmitObsoleteFiles) {
// Regression test for race condition where an obsolete file is returned to
// user as a "live file" but then deleted, all while file deletions are
// disabled.
//
// It happened like this:
//
// 1. [flush thread] Log file "x.log" found by FindObsoleteFiles
// 2. [user thread] DisableFileDeletions, GetSortedWalFiles are called and the
// latter returned "x.log"
// 3. [flush thread] PurgeObsoleteFiles deleted "x.log"
// 4. [user thread] Reading "x.log" failed
//
// Unfortunately the only regression test I can come up with involves sleep.
// We cannot set SyncPoints to repro since, once the fix is applied, the
// SyncPoints would cause a deadlock as the repro's sequence of events is now
// prohibited.
//
// Instead, if we sleep for a second between Find and Purge, and ensure the
// read attempt happens after purge, then the sequence of events will almost
// certainly happen on the old code.
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->LoadDependency({
{"DBImpl::BackgroundCallFlush:FilesFound",
"DBTest2::LiveFilesOmitObsoleteFiles:FlushTriggered"},
{"DBImpl::PurgeObsoleteFiles:End",
"DBTest2::LiveFilesOmitObsoleteFiles:LiveFilesCaptured"},
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"DBImpl::PurgeObsoleteFiles:Begin",
[&](void* /*arg*/) { env_->SleepForMicroseconds(1000000); });
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
ASSERT_OK(Put("key", "val"));
FlushOptions flush_opts;
flush_opts.wait = false;
db_->Flush(flush_opts);
TEST_SYNC_POINT("DBTest2::LiveFilesOmitObsoleteFiles:FlushTriggered");
ASSERT_OK(db_->DisableFileDeletions());
VectorLogPtr log_files;
ASSERT_OK(db_->GetSortedWalFiles(log_files));
TEST_SYNC_POINT("DBTest2::LiveFilesOmitObsoleteFiles:LiveFilesCaptured");
for (const auto& log_file : log_files) {
ASSERT_OK(env_->FileExists(LogFileName(dbname_, log_file->LogNumber())));
}
ASSERT_OK(db_->EnableFileDeletions());
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, TestNumPread) {
Options options = CurrentOptions();
Create a CustomEnv class; Add WinFileSystem; Make LegacyFileSystemWrapper private (#7703) Summary: This PR does the following: -> Creates a WinFileSystem class. This class is the Windows equivalent of the PosixFileSystem and will be used on Windows systems. -> Introduces a CustomEnv class. A CustomEnv is an Env that takes a FileSystem as constructor argument. I believe there will only ever be two implementations of this class (PosixEnv and WinEnv). There is still a CustomEnvWrapper class that takes an Env and a FileSystem and wraps the Env calls with the input Env but uses the FileSystem for the FileSystem calls -> Eliminates the public uses of the LegacyFileSystemWrapper. With this change in place, there are effectively the following patterns of Env: - "Base Env classes" (PosixEnv, WinEnv). These classes implement the core Env functions (e.g. Threads) and have a hard-coded input FileSystem. These classes inherit from CompositeEnv, implement the core Env functions (threads) and delegate the FileSystem-like calls to the input file system. - Wrapped Composite Env classes (MemEnv). These classes take in an Env and a FileSystem. The core env functions are re-directed to the wrapped env. The file system calls are redirected to the input file system - Legacy Wrapped Env classes. These classes take in an Env input (but no FileSystem). The core env functions are re-directed to the wrapped env. A "Legacy File System" is created using this env and the file system calls directed to the env itself. With these changes in place, the PosixEnv becomes a singleton -- there is only ever one created. Any other use of the PosixEnv is via another wrapped env. This cleans up some of the issues with the env construction and destruction. Additionally, there were places in the code that required had an Env when they required a FileSystem. Many of these places would wrap the Env with a LegacyFileSystemWrapper instead of using the env->GetFileSystem(). These places were changed, thereby removing layers of additional redirection (LegacyFileSystem --> Env --> Env::FileSystem). Pull Request resolved: https://github.com/facebook/rocksdb/pull/7703 Reviewed By: zhichao-cao Differential Revision: D25762190 Pulled By: anand1976 fbshipit-source-id: 1a088e97fc916f28ac69c149cd1dcad0ab31704b
4 years ago
bool prefetch_supported =
test::IsPrefetchSupported(env_->GetFileSystem(), dbname_);
// disable block cache
BlockBasedTableOptions table_options;
table_options.no_block_cache = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
Reopen(options);
env_->count_random_reads_ = true;
env_->random_file_open_counter_.store(0);
ASSERT_OK(Put("bar", "foo"));
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Flush());
Create a CustomEnv class; Add WinFileSystem; Make LegacyFileSystemWrapper private (#7703) Summary: This PR does the following: -> Creates a WinFileSystem class. This class is the Windows equivalent of the PosixFileSystem and will be used on Windows systems. -> Introduces a CustomEnv class. A CustomEnv is an Env that takes a FileSystem as constructor argument. I believe there will only ever be two implementations of this class (PosixEnv and WinEnv). There is still a CustomEnvWrapper class that takes an Env and a FileSystem and wraps the Env calls with the input Env but uses the FileSystem for the FileSystem calls -> Eliminates the public uses of the LegacyFileSystemWrapper. With this change in place, there are effectively the following patterns of Env: - "Base Env classes" (PosixEnv, WinEnv). These classes implement the core Env functions (e.g. Threads) and have a hard-coded input FileSystem. These classes inherit from CompositeEnv, implement the core Env functions (threads) and delegate the FileSystem-like calls to the input file system. - Wrapped Composite Env classes (MemEnv). These classes take in an Env and a FileSystem. The core env functions are re-directed to the wrapped env. The file system calls are redirected to the input file system - Legacy Wrapped Env classes. These classes take in an Env input (but no FileSystem). The core env functions are re-directed to the wrapped env. A "Legacy File System" is created using this env and the file system calls directed to the env itself. With these changes in place, the PosixEnv becomes a singleton -- there is only ever one created. Any other use of the PosixEnv is via another wrapped env. This cleans up some of the issues with the env construction and destruction. Additionally, there were places in the code that required had an Env when they required a FileSystem. Many of these places would wrap the Env with a LegacyFileSystemWrapper instead of using the env->GetFileSystem(). These places were changed, thereby removing layers of additional redirection (LegacyFileSystem --> Env --> Env::FileSystem). Pull Request resolved: https://github.com/facebook/rocksdb/pull/7703 Reviewed By: zhichao-cao Differential Revision: D25762190 Pulled By: anand1976 fbshipit-source-id: 1a088e97fc916f28ac69c149cd1dcad0ab31704b
4 years ago
if (prefetch_supported) {
// After flush, we'll open the file and read footer, meta block,
// property block and index block.
ASSERT_EQ(4, env_->random_read_counter_.Read());
} else {
// With prefetch not supported, we will do a single read into a buffer
ASSERT_EQ(1, env_->random_read_counter_.Read());
}
ASSERT_EQ(1, env_->random_file_open_counter_.load());
// One pread per a normal data block read
env_->random_file_open_counter_.store(0);
env_->random_read_counter_.Reset();
ASSERT_EQ("bar", Get("foo"));
ASSERT_EQ(1, env_->random_read_counter_.Read());
// All files are already opened.
ASSERT_EQ(0, env_->random_file_open_counter_.load());
env_->random_file_open_counter_.store(0);
env_->random_read_counter_.Reset();
ASSERT_OK(Put("bar2", "foo2"));
ASSERT_OK(Put("foo2", "bar2"));
ASSERT_OK(Flush());
Create a CustomEnv class; Add WinFileSystem; Make LegacyFileSystemWrapper private (#7703) Summary: This PR does the following: -> Creates a WinFileSystem class. This class is the Windows equivalent of the PosixFileSystem and will be used on Windows systems. -> Introduces a CustomEnv class. A CustomEnv is an Env that takes a FileSystem as constructor argument. I believe there will only ever be two implementations of this class (PosixEnv and WinEnv). There is still a CustomEnvWrapper class that takes an Env and a FileSystem and wraps the Env calls with the input Env but uses the FileSystem for the FileSystem calls -> Eliminates the public uses of the LegacyFileSystemWrapper. With this change in place, there are effectively the following patterns of Env: - "Base Env classes" (PosixEnv, WinEnv). These classes implement the core Env functions (e.g. Threads) and have a hard-coded input FileSystem. These classes inherit from CompositeEnv, implement the core Env functions (threads) and delegate the FileSystem-like calls to the input file system. - Wrapped Composite Env classes (MemEnv). These classes take in an Env and a FileSystem. The core env functions are re-directed to the wrapped env. The file system calls are redirected to the input file system - Legacy Wrapped Env classes. These classes take in an Env input (but no FileSystem). The core env functions are re-directed to the wrapped env. A "Legacy File System" is created using this env and the file system calls directed to the env itself. With these changes in place, the PosixEnv becomes a singleton -- there is only ever one created. Any other use of the PosixEnv is via another wrapped env. This cleans up some of the issues with the env construction and destruction. Additionally, there were places in the code that required had an Env when they required a FileSystem. Many of these places would wrap the Env with a LegacyFileSystemWrapper instead of using the env->GetFileSystem(). These places were changed, thereby removing layers of additional redirection (LegacyFileSystem --> Env --> Env::FileSystem). Pull Request resolved: https://github.com/facebook/rocksdb/pull/7703 Reviewed By: zhichao-cao Differential Revision: D25762190 Pulled By: anand1976 fbshipit-source-id: 1a088e97fc916f28ac69c149cd1dcad0ab31704b
4 years ago
if (prefetch_supported) {
// After flush, we'll open the file and read footer, meta block,
// property block and index block.
ASSERT_EQ(4, env_->random_read_counter_.Read());
} else {
// With prefetch not supported, we will do a single read into a buffer
ASSERT_EQ(1, env_->random_read_counter_.Read());
}
ASSERT_EQ(1, env_->random_file_open_counter_.load());
env_->random_file_open_counter_.store(0);
env_->random_read_counter_.Reset();
ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
Create a CustomEnv class; Add WinFileSystem; Make LegacyFileSystemWrapper private (#7703) Summary: This PR does the following: -> Creates a WinFileSystem class. This class is the Windows equivalent of the PosixFileSystem and will be used on Windows systems. -> Introduces a CustomEnv class. A CustomEnv is an Env that takes a FileSystem as constructor argument. I believe there will only ever be two implementations of this class (PosixEnv and WinEnv). There is still a CustomEnvWrapper class that takes an Env and a FileSystem and wraps the Env calls with the input Env but uses the FileSystem for the FileSystem calls -> Eliminates the public uses of the LegacyFileSystemWrapper. With this change in place, there are effectively the following patterns of Env: - "Base Env classes" (PosixEnv, WinEnv). These classes implement the core Env functions (e.g. Threads) and have a hard-coded input FileSystem. These classes inherit from CompositeEnv, implement the core Env functions (threads) and delegate the FileSystem-like calls to the input file system. - Wrapped Composite Env classes (MemEnv). These classes take in an Env and a FileSystem. The core env functions are re-directed to the wrapped env. The file system calls are redirected to the input file system - Legacy Wrapped Env classes. These classes take in an Env input (but no FileSystem). The core env functions are re-directed to the wrapped env. A "Legacy File System" is created using this env and the file system calls directed to the env itself. With these changes in place, the PosixEnv becomes a singleton -- there is only ever one created. Any other use of the PosixEnv is via another wrapped env. This cleans up some of the issues with the env construction and destruction. Additionally, there were places in the code that required had an Env when they required a FileSystem. Many of these places would wrap the Env with a LegacyFileSystemWrapper instead of using the env->GetFileSystem(). These places were changed, thereby removing layers of additional redirection (LegacyFileSystem --> Env --> Env::FileSystem). Pull Request resolved: https://github.com/facebook/rocksdb/pull/7703 Reviewed By: zhichao-cao Differential Revision: D25762190 Pulled By: anand1976 fbshipit-source-id: 1a088e97fc916f28ac69c149cd1dcad0ab31704b
4 years ago
if (prefetch_supported) {
// Compaction needs two input blocks, which requires 2 preads, and
// generate a new SST file which needs 4 preads (footer, meta block,
// property block and index block). In total 6.
ASSERT_EQ(6, env_->random_read_counter_.Read());
} else {
// With prefetch off, compaction needs two input blocks,
// followed by a single buffered read. In total 3.
ASSERT_EQ(3, env_->random_read_counter_.Read());
}
// All compaction input files should have already been opened.
ASSERT_EQ(1, env_->random_file_open_counter_.load());
// One pread per a normal data block read
env_->random_file_open_counter_.store(0);
env_->random_read_counter_.Reset();
ASSERT_EQ("foo2", Get("bar2"));
ASSERT_EQ(1, env_->random_read_counter_.Read());
// SST files are already opened.
ASSERT_EQ(0, env_->random_file_open_counter_.load());
}
class TraceExecutionResultHandler : public TraceRecordResult::Handler {
public:
TraceExecutionResultHandler() {}
~TraceExecutionResultHandler() override {}
virtual Status Handle(const StatusOnlyTraceExecutionResult& result) override {
if (result.GetStartTimestamp() > result.GetEndTimestamp()) {
return Status::InvalidArgument("Invalid timestamps.");
}
result.GetStatus().PermitUncheckedError();
switch (result.GetTraceType()) {
case kTraceWrite: {
total_latency_ += result.GetLatency();
cnt_++;
writes_++;
break;
}
default:
return Status::Corruption("Type mismatch.");
}
return Status::OK();
}
virtual Status Handle(
const SingleValueTraceExecutionResult& result) override {
if (result.GetStartTimestamp() > result.GetEndTimestamp()) {
return Status::InvalidArgument("Invalid timestamps.");
}
result.GetStatus().PermitUncheckedError();
switch (result.GetTraceType()) {
case kTraceGet: {
total_latency_ += result.GetLatency();
cnt_++;
gets_++;
break;
}
default:
return Status::Corruption("Type mismatch.");
}
return Status::OK();
}
virtual Status Handle(
const MultiValuesTraceExecutionResult& result) override {
if (result.GetStartTimestamp() > result.GetEndTimestamp()) {
return Status::InvalidArgument("Invalid timestamps.");
}
for (const Status& s : result.GetMultiStatus()) {
s.PermitUncheckedError();
}
switch (result.GetTraceType()) {
case kTraceMultiGet: {
total_latency_ += result.GetLatency();
cnt_++;
multigets_++;
break;
}
default:
return Status::Corruption("Type mismatch.");
}
return Status::OK();
}
virtual Status Handle(const IteratorTraceExecutionResult& result) override {
if (result.GetStartTimestamp() > result.GetEndTimestamp()) {
return Status::InvalidArgument("Invalid timestamps.");
}
result.GetStatus().PermitUncheckedError();
switch (result.GetTraceType()) {
case kTraceIteratorSeek:
case kTraceIteratorSeekForPrev: {
total_latency_ += result.GetLatency();
cnt_++;
seeks_++;
break;
}
default:
return Status::Corruption("Type mismatch.");
}
return Status::OK();
}
void Reset() {
total_latency_ = 0;
cnt_ = 0;
writes_ = 0;
gets_ = 0;
seeks_ = 0;
multigets_ = 0;
}
double GetAvgLatency() const {
return cnt_ == 0 ? 0.0 : 1.0 * total_latency_ / cnt_;
}
int GetNumWrites() const { return writes_; }
int GetNumGets() const { return gets_; }
int GetNumIterSeeks() const { return seeks_; }
int GetNumMultiGets() const { return multigets_; }
private:
std::atomic<uint64_t> total_latency_{0};
std::atomic<uint32_t> cnt_{0};
std::atomic<int> writes_{0};
std::atomic<int> gets_{0};
std::atomic<int> seeks_{0};
std::atomic<int> multigets_{0};
};
TEST_F(DBTest2, TraceAndReplay) {
Options options = CurrentOptions();
options.merge_operator = MergeOperators::CreatePutOperator();
ReadOptions ro;
WriteOptions wo;
TraceOptions trace_opts;
EnvOptions env_opts;
CreateAndReopenWithCF({"pikachu"}, options);
Random rnd(301);
Iterator* single_iter = nullptr;
ASSERT_TRUE(db_->EndTrace().IsIOError());
std::string trace_filename = dbname_ + "/rocksdb.trace";
std::unique_ptr<TraceWriter> trace_writer;
ASSERT_OK(NewFileTraceWriter(env_, env_opts, trace_filename, &trace_writer));
ASSERT_OK(db_->StartTrace(trace_opts, std::move(trace_writer)));
// 5 Writes
ASSERT_OK(Put(0, "a", "1"));
ASSERT_OK(Merge(0, "b", "2"));
ASSERT_OK(Delete(0, "c"));
ASSERT_OK(SingleDelete(0, "d"));
ASSERT_OK(db_->DeleteRange(wo, dbfull()->DefaultColumnFamily(), "e", "f"));
// 6th Write
WriteBatch batch;
ASSERT_OK(batch.Put("f", "11"));
ASSERT_OK(batch.Merge("g", "12"));
ASSERT_OK(batch.Delete("h"));
ASSERT_OK(batch.SingleDelete("i"));
ASSERT_OK(batch.DeleteRange("j", "k"));
ASSERT_OK(db_->Write(wo, &batch));
// 2 Seek(ForPrev)s
single_iter = db_->NewIterator(ro);
single_iter->Seek("f"); // Seek 1
single_iter->SeekForPrev("g");
ASSERT_OK(single_iter->status());
delete single_iter;
// 2 Gets
ASSERT_EQ("1", Get(0, "a"));
ASSERT_EQ("12", Get(0, "g"));
// 7th and 8th Write, 3rd Get
ASSERT_OK(Put(1, "foo", "bar"));
ASSERT_OK(Put(1, "rocksdb", "rocks"));
ASSERT_EQ("NOT_FOUND", Get(1, "leveldb"));
// Total Write x 8, Get x 3, Seek x 2.
ASSERT_OK(db_->EndTrace());
// These should not get into the trace file as it is after EndTrace.
ASSERT_OK(Put("hello", "world"));
ASSERT_OK(Merge("foo", "bar"));
// Open another db, replay, and verify the data
std::string value;
std::string dbname2 = test::PerThreadDBPath(env_, "/db_replay");
ASSERT_OK(DestroyDB(dbname2, options));
// Using a different name than db2, to pacify infer's use-after-lifetime
// warnings (http://fbinfer.com).
DB* db2_init = nullptr;
options.create_if_missing = true;
ASSERT_OK(DB::Open(options, dbname2, &db2_init));
ColumnFamilyHandle* cf;
ASSERT_OK(
db2_init->CreateColumnFamily(ColumnFamilyOptions(), "pikachu", &cf));
delete cf;
delete db2_init;
DB* db2 = nullptr;
std::vector<ColumnFamilyDescriptor> column_families;
ColumnFamilyOptions cf_options;
cf_options.merge_operator = MergeOperators::CreatePutOperator();
column_families.push_back(ColumnFamilyDescriptor("default", cf_options));
column_families.push_back(
ColumnFamilyDescriptor("pikachu", ColumnFamilyOptions()));
std::vector<ColumnFamilyHandle*> handles;
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
DBOptions db_opts;
db_opts.env = env_;
ASSERT_OK(DB::Open(db_opts, dbname2, column_families, &handles, &db2));
env_->SleepForMicroseconds(100);
// Verify that the keys don't already exist
ASSERT_TRUE(db2->Get(ro, handles[0], "a", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "g", &value).IsNotFound());
std::unique_ptr<TraceReader> trace_reader;
ASSERT_OK(NewFileTraceReader(env_, env_opts, trace_filename, &trace_reader));
std::unique_ptr<Replayer> replayer;
ASSERT_OK(
db2->NewDefaultReplayer(handles, std::move(trace_reader), &replayer));
TraceExecutionResultHandler res_handler;
std::function<void(Status, std::unique_ptr<TraceRecordResult> &&)> res_cb =
[&res_handler](Status exec_s, std::unique_ptr<TraceRecordResult>&& res) {
ASSERT_TRUE(exec_s.ok() || exec_s.IsNotSupported());
if (res != nullptr) {
ASSERT_OK(res->Accept(&res_handler));
res.reset();
}
};
// Unprepared replay should fail with Status::Incomplete()
ASSERT_TRUE(replayer->Replay(ReplayOptions(), nullptr).IsIncomplete());
ASSERT_OK(replayer->Prepare());
// Ok to repeatedly Prepare().
ASSERT_OK(replayer->Prepare());
// Replay using 1 thread, 1x speed.
ASSERT_OK(replayer->Replay(ReplayOptions(1, 1.0), res_cb));
ASSERT_GT(res_handler.GetAvgLatency(), 0.0);
ASSERT_EQ(res_handler.GetNumWrites(), 8);
ASSERT_EQ(res_handler.GetNumGets(), 3);
ASSERT_EQ(res_handler.GetNumIterSeeks(), 2);
ASSERT_EQ(res_handler.GetNumMultiGets(), 0);
res_handler.Reset();
ASSERT_OK(db2->Get(ro, handles[0], "a", &value));
ASSERT_EQ("1", value);
ASSERT_OK(db2->Get(ro, handles[0], "g", &value));
ASSERT_EQ("12", value);
ASSERT_TRUE(db2->Get(ro, handles[0], "hello", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "world", &value).IsNotFound());
ASSERT_OK(db2->Get(ro, handles[1], "foo", &value));
ASSERT_EQ("bar", value);
ASSERT_OK(db2->Get(ro, handles[1], "rocksdb", &value));
ASSERT_EQ("rocks", value);
// Re-replay should fail with Status::Incomplete() if Prepare() was not
// called. Currently we don't distinguish between unprepared and trace end.
ASSERT_TRUE(replayer->Replay(ReplayOptions(), nullptr).IsIncomplete());
// Re-replay using 2 threads, 2x speed.
ASSERT_OK(replayer->Prepare());
ASSERT_OK(replayer->Replay(ReplayOptions(2, 2.0), res_cb));
ASSERT_GT(res_handler.GetAvgLatency(), 0.0);
ASSERT_EQ(res_handler.GetNumWrites(), 8);
ASSERT_EQ(res_handler.GetNumGets(), 3);
ASSERT_EQ(res_handler.GetNumIterSeeks(), 2);
ASSERT_EQ(res_handler.GetNumMultiGets(), 0);
res_handler.Reset();
// Re-replay using 2 threads, 1/2 speed.
ASSERT_OK(replayer->Prepare());
ASSERT_OK(replayer->Replay(ReplayOptions(2, 0.5), res_cb));
ASSERT_GT(res_handler.GetAvgLatency(), 0.0);
ASSERT_EQ(res_handler.GetNumWrites(), 8);
ASSERT_EQ(res_handler.GetNumGets(), 3);
ASSERT_EQ(res_handler.GetNumIterSeeks(), 2);
ASSERT_EQ(res_handler.GetNumMultiGets(), 0);
res_handler.Reset();
replayer.reset();
for (auto handle : handles) {
delete handle;
}
delete db2;
ASSERT_OK(DestroyDB(dbname2, options));
}
TEST_F(DBTest2, TraceAndManualReplay) {
Options options = CurrentOptions();
options.merge_operator = MergeOperators::CreatePutOperator();
ReadOptions ro;
WriteOptions wo;
TraceOptions trace_opts;
EnvOptions env_opts;
CreateAndReopenWithCF({"pikachu"}, options);
Random rnd(301);
Iterator* single_iter = nullptr;
ASSERT_TRUE(db_->EndTrace().IsIOError());
std::string trace_filename = dbname_ + "/rocksdb.trace";
std::unique_ptr<TraceWriter> trace_writer;
ASSERT_OK(NewFileTraceWriter(env_, env_opts, trace_filename, &trace_writer));
ASSERT_OK(db_->StartTrace(trace_opts, std::move(trace_writer)));
ASSERT_OK(Put(0, "a", "1"));
ASSERT_OK(Merge(0, "b", "2"));
ASSERT_OK(Delete(0, "c"));
ASSERT_OK(SingleDelete(0, "d"));
ASSERT_OK(db_->DeleteRange(wo, dbfull()->DefaultColumnFamily(), "e", "f"));
WriteBatch batch;
ASSERT_OK(batch.Put("f", "11"));
ASSERT_OK(batch.Merge("g", "12"));
ASSERT_OK(batch.Delete("h"));
ASSERT_OK(batch.SingleDelete("i"));
ASSERT_OK(batch.DeleteRange("j", "k"));
ASSERT_OK(db_->Write(wo, &batch));
single_iter = db_->NewIterator(ro);
single_iter->Seek("f");
single_iter->SeekForPrev("g");
ASSERT_OK(single_iter->status());
delete single_iter;
// Write some sequenced keys for testing lower/upper bounds of iterator.
batch.Clear();
ASSERT_OK(batch.Put("iter-0", "iter-0"));
ASSERT_OK(batch.Put("iter-1", "iter-1"));
ASSERT_OK(batch.Put("iter-2", "iter-2"));
ASSERT_OK(batch.Put("iter-3", "iter-3"));
ASSERT_OK(batch.Put("iter-4", "iter-4"));
ASSERT_OK(db_->Write(wo, &batch));
ReadOptions bounded_ro = ro;
Slice lower_bound("iter-1");
Slice upper_bound("iter-3");
bounded_ro.iterate_lower_bound = &lower_bound;
bounded_ro.iterate_upper_bound = &upper_bound;
single_iter = db_->NewIterator(bounded_ro);
single_iter->Seek("iter-0");
ASSERT_EQ(single_iter->key().ToString(), "iter-1");
single_iter->Seek("iter-2");
ASSERT_EQ(single_iter->key().ToString(), "iter-2");
single_iter->Seek("iter-4");
ASSERT_FALSE(single_iter->Valid());
single_iter->SeekForPrev("iter-0");
ASSERT_FALSE(single_iter->Valid());
single_iter->SeekForPrev("iter-2");
ASSERT_EQ(single_iter->key().ToString(), "iter-2");
single_iter->SeekForPrev("iter-4");
ASSERT_EQ(single_iter->key().ToString(), "iter-2");
ASSERT_OK(single_iter->status());
delete single_iter;
ASSERT_EQ("1", Get(0, "a"));
ASSERT_EQ("12", Get(0, "g"));
ASSERT_OK(Put(1, "foo", "bar"));
ASSERT_OK(Put(1, "rocksdb", "rocks"));
ASSERT_EQ("NOT_FOUND", Get(1, "leveldb"));
// Same as TraceAndReplay, Write x 8, Get x 3, Seek x 2.
// Plus 1 WriteBatch for iterator with lower/upper bounds, and 6
// Seek(ForPrev)s.
// Total Write x 9, Get x 3, Seek x 8
ASSERT_OK(db_->EndTrace());
// These should not get into the trace file as it is after EndTrace.
ASSERT_OK(Put("hello", "world"));
ASSERT_OK(Merge("foo", "bar"));
// Open another db, replay, and verify the data
std::string value;
std::string dbname2 = test::PerThreadDBPath(env_, "/db_replay");
ASSERT_OK(DestroyDB(dbname2, options));
// Using a different name than db2, to pacify infer's use-after-lifetime
// warnings (http://fbinfer.com).
DB* db2_init = nullptr;
options.create_if_missing = true;
ASSERT_OK(DB::Open(options, dbname2, &db2_init));
ColumnFamilyHandle* cf;
ASSERT_OK(
db2_init->CreateColumnFamily(ColumnFamilyOptions(), "pikachu", &cf));
delete cf;
delete db2_init;
DB* db2 = nullptr;
std::vector<ColumnFamilyDescriptor> column_families;
ColumnFamilyOptions cf_options;
cf_options.merge_operator = MergeOperators::CreatePutOperator();
column_families.push_back(ColumnFamilyDescriptor("default", cf_options));
column_families.push_back(
ColumnFamilyDescriptor("pikachu", ColumnFamilyOptions()));
std::vector<ColumnFamilyHandle*> handles;
DBOptions db_opts;
db_opts.env = env_;
ASSERT_OK(DB::Open(db_opts, dbname2, column_families, &handles, &db2));
env_->SleepForMicroseconds(100);
// Verify that the keys don't already exist
ASSERT_TRUE(db2->Get(ro, handles[0], "a", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "g", &value).IsNotFound());
std::unique_ptr<TraceReader> trace_reader;
ASSERT_OK(NewFileTraceReader(env_, env_opts, trace_filename, &trace_reader));
std::unique_ptr<Replayer> replayer;
ASSERT_OK(
db2->NewDefaultReplayer(handles, std::move(trace_reader), &replayer));
TraceExecutionResultHandler res_handler;
// Manual replay for 2 times. The 2nd checks if the replay can restart.
std::unique_ptr<TraceRecord> record;
std::unique_ptr<TraceRecordResult> result;
for (int i = 0; i < 2; i++) {
// Next should fail if unprepared.
ASSERT_TRUE(replayer->Next(nullptr).IsIncomplete());
ASSERT_OK(replayer->Prepare());
Status s = Status::OK();
// Looping until trace end.
while (s.ok()) {
s = replayer->Next(&record);
// Skip unsupported operations.
if (s.IsNotSupported()) {
continue;
}
if (s.ok()) {
ASSERT_OK(replayer->Execute(record, &result));
if (result != nullptr) {
ASSERT_OK(result->Accept(&res_handler));
if (record->GetTraceType() == kTraceIteratorSeek ||
record->GetTraceType() == kTraceIteratorSeekForPrev) {
IteratorSeekQueryTraceRecord* iter_rec =
dynamic_cast<IteratorSeekQueryTraceRecord*>(record.get());
IteratorTraceExecutionResult* iter_res =
dynamic_cast<IteratorTraceExecutionResult*>(result.get());
// Check if lower/upper bounds are correctly saved and decoded.
std::string lower_str = iter_rec->GetLowerBound().ToString();
std::string upper_str = iter_rec->GetUpperBound().ToString();
std::string iter_key = iter_res->GetKey().ToString();
std::string iter_value = iter_res->GetValue().ToString();
if (!lower_str.empty() && !upper_str.empty()) {
ASSERT_EQ(lower_str, "iter-1");
ASSERT_EQ(upper_str, "iter-3");
if (iter_res->GetValid()) {
// If iterator is valid, then lower_bound <= key < upper_bound.
ASSERT_GE(iter_key, lower_str);
ASSERT_LT(iter_key, upper_str);
} else {
// If iterator is invalid, then
// key < lower_bound or key >= upper_bound.
ASSERT_TRUE(iter_key < lower_str || iter_key >= upper_str);
}
}
// If iterator is invalid, the key and value should be empty.
if (!iter_res->GetValid()) {
ASSERT_TRUE(iter_key.empty());
ASSERT_TRUE(iter_value.empty());
}
}
result.reset();
}
}
}
// Status::Incomplete() will be returned when manually reading the trace
// end, or Prepare() was not called.
ASSERT_TRUE(s.IsIncomplete());
ASSERT_TRUE(replayer->Next(nullptr).IsIncomplete());
ASSERT_GT(res_handler.GetAvgLatency(), 0.0);
ASSERT_EQ(res_handler.GetNumWrites(), 9);
ASSERT_EQ(res_handler.GetNumGets(), 3);
ASSERT_EQ(res_handler.GetNumIterSeeks(), 8);
ASSERT_EQ(res_handler.GetNumMultiGets(), 0);
res_handler.Reset();
}
ASSERT_OK(db2->Get(ro, handles[0], "a", &value));
ASSERT_EQ("1", value);
ASSERT_OK(db2->Get(ro, handles[0], "g", &value));
ASSERT_EQ("12", value);
ASSERT_TRUE(db2->Get(ro, handles[0], "hello", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "world", &value).IsNotFound());
ASSERT_OK(db2->Get(ro, handles[1], "foo", &value));
ASSERT_EQ("bar", value);
ASSERT_OK(db2->Get(ro, handles[1], "rocksdb", &value));
ASSERT_EQ("rocks", value);
// Test execution of artificially created TraceRecords.
uint64_t fake_ts = 1U;
// Write
batch.Clear();
ASSERT_OK(batch.Put("trace-record-write1", "write1"));
ASSERT_OK(batch.Put("trace-record-write2", "write2"));
record.reset(new WriteQueryTraceRecord(batch.Data(), fake_ts++));
ASSERT_OK(replayer->Execute(record, &result));
ASSERT_TRUE(result != nullptr);
ASSERT_OK(result->Accept(&res_handler)); // Write x 1
ASSERT_OK(db2->Get(ro, handles[0], "trace-record-write1", &value));
ASSERT_EQ("write1", value);
ASSERT_OK(db2->Get(ro, handles[0], "trace-record-write2", &value));
ASSERT_EQ("write2", value);
ASSERT_GT(res_handler.GetAvgLatency(), 0.0);
ASSERT_EQ(res_handler.GetNumWrites(), 1);
ASSERT_EQ(res_handler.GetNumGets(), 0);
ASSERT_EQ(res_handler.GetNumIterSeeks(), 0);
ASSERT_EQ(res_handler.GetNumMultiGets(), 0);
res_handler.Reset();
// Get related
// Get an existing key.
record.reset(new GetQueryTraceRecord(handles[0]->GetID(),
"trace-record-write1", fake_ts++));
ASSERT_OK(replayer->Execute(record, &result));
ASSERT_TRUE(result != nullptr);
ASSERT_OK(result->Accept(&res_handler)); // Get x 1
// Get an non-existing key, should still return Status::OK().
record.reset(new GetQueryTraceRecord(handles[0]->GetID(), "trace-record-get",
fake_ts++));
ASSERT_OK(replayer->Execute(record, &result));
ASSERT_TRUE(result != nullptr);
ASSERT_OK(result->Accept(&res_handler)); // Get x 2
// Get from an invalid (non-existing) cf_id.
uint32_t invalid_cf_id = handles[1]->GetID() + 1;
record.reset(new GetQueryTraceRecord(invalid_cf_id, "whatever", fake_ts++));
ASSERT_TRUE(replayer->Execute(record, &result).IsCorruption());
ASSERT_TRUE(result == nullptr);
ASSERT_GT(res_handler.GetAvgLatency(), 0.0);
ASSERT_EQ(res_handler.GetNumWrites(), 0);
ASSERT_EQ(res_handler.GetNumGets(), 2);
ASSERT_EQ(res_handler.GetNumIterSeeks(), 0);
ASSERT_EQ(res_handler.GetNumMultiGets(), 0);
res_handler.Reset();
// Iteration related
for (IteratorSeekQueryTraceRecord::SeekType seekType :
{IteratorSeekQueryTraceRecord::kSeek,
IteratorSeekQueryTraceRecord::kSeekForPrev}) {
// Seek to an existing key.
record.reset(new IteratorSeekQueryTraceRecord(
seekType, handles[0]->GetID(), "trace-record-write1", fake_ts++));
ASSERT_OK(replayer->Execute(record, &result));
ASSERT_TRUE(result != nullptr);
ASSERT_OK(result->Accept(&res_handler)); // Seek x 1 in one iteration
// Seek to an non-existing key, should still return Status::OK().
record.reset(new IteratorSeekQueryTraceRecord(
seekType, handles[0]->GetID(), "trace-record-get", fake_ts++));
ASSERT_OK(replayer->Execute(record, &result));
ASSERT_TRUE(result != nullptr);
ASSERT_OK(result->Accept(&res_handler)); // Seek x 2 in one iteration
// Seek from an invalid cf_id.
record.reset(new IteratorSeekQueryTraceRecord(seekType, invalid_cf_id,
"whatever", fake_ts++));
ASSERT_TRUE(replayer->Execute(record, &result).IsCorruption());
ASSERT_TRUE(result == nullptr);
}
ASSERT_GT(res_handler.GetAvgLatency(), 0.0);
ASSERT_EQ(res_handler.GetNumWrites(), 0);
ASSERT_EQ(res_handler.GetNumGets(), 0);
ASSERT_EQ(res_handler.GetNumIterSeeks(), 4); // Seek x 2 in two iterations
ASSERT_EQ(res_handler.GetNumMultiGets(), 0);
res_handler.Reset();
// MultiGet related
// Get existing keys.
record.reset(new MultiGetQueryTraceRecord(
std::vector<uint32_t>({handles[0]->GetID(), handles[1]->GetID()}),
std::vector<std::string>({"a", "foo"}), fake_ts++));
ASSERT_OK(replayer->Execute(record, &result));
ASSERT_TRUE(result != nullptr);
ASSERT_OK(result->Accept(&res_handler)); // MultiGet x 1
// Get all non-existing keys, should still return Status::OK().
record.reset(new MultiGetQueryTraceRecord(
std::vector<uint32_t>({handles[0]->GetID(), handles[1]->GetID()}),
std::vector<std::string>({"no1", "no2"}), fake_ts++));
ASSERT_OK(replayer->Execute(record, &result));
ASSERT_TRUE(result != nullptr);
ASSERT_OK(result->Accept(&res_handler)); // MultiGet x 2
// Get mixed of existing and non-existing keys, should still return
// Status::OK().
record.reset(new MultiGetQueryTraceRecord(
std::vector<uint32_t>({handles[0]->GetID(), handles[1]->GetID()}),
std::vector<std::string>({"a", "no2"}), fake_ts++));
ASSERT_OK(replayer->Execute(record, &result));
ASSERT_TRUE(result != nullptr);
MultiValuesTraceExecutionResult* mvr =
dynamic_cast<MultiValuesTraceExecutionResult*>(result.get());
ASSERT_TRUE(mvr != nullptr);
ASSERT_OK(mvr->GetMultiStatus()[0]);
ASSERT_TRUE(mvr->GetMultiStatus()[1].IsNotFound());
ASSERT_EQ(mvr->GetValues()[0], "1");
ASSERT_EQ(mvr->GetValues()[1], "");
ASSERT_OK(result->Accept(&res_handler)); // MultiGet x 3
// Get from an invalid (non-existing) cf_id.
record.reset(new MultiGetQueryTraceRecord(
std::vector<uint32_t>(
{handles[0]->GetID(), handles[1]->GetID(), invalid_cf_id}),
std::vector<std::string>({"a", "foo", "whatever"}), fake_ts++));
ASSERT_TRUE(replayer->Execute(record, &result).IsCorruption());
ASSERT_TRUE(result == nullptr);
// Empty MultiGet
record.reset(new MultiGetQueryTraceRecord(
std::vector<uint32_t>(), std::vector<std::string>(), fake_ts++));
ASSERT_TRUE(replayer->Execute(record, &result).IsInvalidArgument());
ASSERT_TRUE(result == nullptr);
// MultiGet size mismatch
record.reset(new MultiGetQueryTraceRecord(
std::vector<uint32_t>({handles[0]->GetID(), handles[1]->GetID()}),
std::vector<std::string>({"a"}), fake_ts++));
ASSERT_TRUE(replayer->Execute(record, &result).IsInvalidArgument());
ASSERT_TRUE(result == nullptr);
ASSERT_GT(res_handler.GetAvgLatency(), 0.0);
ASSERT_EQ(res_handler.GetNumWrites(), 0);
ASSERT_EQ(res_handler.GetNumGets(), 0);
ASSERT_EQ(res_handler.GetNumIterSeeks(), 0);
ASSERT_EQ(res_handler.GetNumMultiGets(), 3);
res_handler.Reset();
replayer.reset();
for (auto handle : handles) {
delete handle;
}
delete db2;
ASSERT_OK(DestroyDB(dbname2, options));
}
TEST_F(DBTest2, TraceWithLimit) {
Options options = CurrentOptions();
options.merge_operator = MergeOperators::CreatePutOperator();
ReadOptions ro;
WriteOptions wo;
TraceOptions trace_opts;
EnvOptions env_opts;
CreateAndReopenWithCF({"pikachu"}, options);
Random rnd(301);
// test the max trace file size options
trace_opts.max_trace_file_size = 5;
std::string trace_filename = dbname_ + "/rocksdb.trace1";
std::unique_ptr<TraceWriter> trace_writer;
ASSERT_OK(NewFileTraceWriter(env_, env_opts, trace_filename, &trace_writer));
ASSERT_OK(db_->StartTrace(trace_opts, std::move(trace_writer)));
ASSERT_OK(Put(0, "a", "1"));
ASSERT_OK(Put(0, "b", "1"));
ASSERT_OK(Put(0, "c", "1"));
ASSERT_OK(db_->EndTrace());
std::string dbname2 = test::PerThreadDBPath(env_, "/db_replay2");
std::string value;
ASSERT_OK(DestroyDB(dbname2, options));
// Using a different name than db2, to pacify infer's use-after-lifetime
// warnings (http://fbinfer.com).
DB* db2_init = nullptr;
options.create_if_missing = true;
ASSERT_OK(DB::Open(options, dbname2, &db2_init));
ColumnFamilyHandle* cf;
ASSERT_OK(
db2_init->CreateColumnFamily(ColumnFamilyOptions(), "pikachu", &cf));
delete cf;
delete db2_init;
DB* db2 = nullptr;
std::vector<ColumnFamilyDescriptor> column_families;
ColumnFamilyOptions cf_options;
cf_options.merge_operator = MergeOperators::CreatePutOperator();
column_families.push_back(ColumnFamilyDescriptor("default", cf_options));
column_families.push_back(
ColumnFamilyDescriptor("pikachu", ColumnFamilyOptions()));
std::vector<ColumnFamilyHandle*> handles;
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
DBOptions db_opts;
db_opts.env = env_;
ASSERT_OK(DB::Open(db_opts, dbname2, column_families, &handles, &db2));
env_->SleepForMicroseconds(100);
// Verify that the keys don't already exist
ASSERT_TRUE(db2->Get(ro, handles[0], "a", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "b", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "c", &value).IsNotFound());
std::unique_ptr<TraceReader> trace_reader;
ASSERT_OK(NewFileTraceReader(env_, env_opts, trace_filename, &trace_reader));
std::unique_ptr<Replayer> replayer;
ASSERT_OK(
db2->NewDefaultReplayer(handles, std::move(trace_reader), &replayer));
ASSERT_OK(replayer->Prepare());
ASSERT_OK(replayer->Replay(ReplayOptions(), nullptr));
replayer.reset();
ASSERT_TRUE(db2->Get(ro, handles[0], "a", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "b", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "c", &value).IsNotFound());
for (auto handle : handles) {
delete handle;
}
delete db2;
ASSERT_OK(DestroyDB(dbname2, options));
}
TEST_F(DBTest2, TraceWithSampling) {
Options options = CurrentOptions();
ReadOptions ro;
WriteOptions wo;
TraceOptions trace_opts;
EnvOptions env_opts;
CreateAndReopenWithCF({"pikachu"}, options);
Random rnd(301);
// test the trace file sampling options
trace_opts.sampling_frequency = 2;
std::string trace_filename = dbname_ + "/rocksdb.trace_sampling";
std::unique_ptr<TraceWriter> trace_writer;
ASSERT_OK(NewFileTraceWriter(env_, env_opts, trace_filename, &trace_writer));
ASSERT_OK(db_->StartTrace(trace_opts, std::move(trace_writer)));
ASSERT_OK(Put(0, "a", "1"));
ASSERT_OK(Put(0, "b", "2"));
ASSERT_OK(Put(0, "c", "3"));
ASSERT_OK(Put(0, "d", "4"));
ASSERT_OK(Put(0, "e", "5"));
ASSERT_OK(db_->EndTrace());
std::string dbname2 = test::PerThreadDBPath(env_, "/db_replay_sampling");
std::string value;
ASSERT_OK(DestroyDB(dbname2, options));
// Using a different name than db2, to pacify infer's use-after-lifetime
// warnings (http://fbinfer.com).
DB* db2_init = nullptr;
options.create_if_missing = true;
ASSERT_OK(DB::Open(options, dbname2, &db2_init));
ColumnFamilyHandle* cf;
ASSERT_OK(
db2_init->CreateColumnFamily(ColumnFamilyOptions(), "pikachu", &cf));
delete cf;
delete db2_init;
DB* db2 = nullptr;
std::vector<ColumnFamilyDescriptor> column_families;
ColumnFamilyOptions cf_options;
column_families.push_back(ColumnFamilyDescriptor("default", cf_options));
column_families.push_back(
ColumnFamilyDescriptor("pikachu", ColumnFamilyOptions()));
std::vector<ColumnFamilyHandle*> handles;
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
DBOptions db_opts;
db_opts.env = env_;
ASSERT_OK(DB::Open(db_opts, dbname2, column_families, &handles, &db2));
env_->SleepForMicroseconds(100);
ASSERT_TRUE(db2->Get(ro, handles[0], "a", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "b", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "c", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "d", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "e", &value).IsNotFound());
std::unique_ptr<TraceReader> trace_reader;
ASSERT_OK(NewFileTraceReader(env_, env_opts, trace_filename, &trace_reader));
std::unique_ptr<Replayer> replayer;
ASSERT_OK(
db2->NewDefaultReplayer(handles, std::move(trace_reader), &replayer));
ASSERT_OK(replayer->Prepare());
ASSERT_OK(replayer->Replay(ReplayOptions(), nullptr));
replayer.reset();
ASSERT_TRUE(db2->Get(ro, handles[0], "a", &value).IsNotFound());
ASSERT_FALSE(db2->Get(ro, handles[0], "b", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "c", &value).IsNotFound());
ASSERT_FALSE(db2->Get(ro, handles[0], "d", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "e", &value).IsNotFound());
for (auto handle : handles) {
delete handle;
}
delete db2;
ASSERT_OK(DestroyDB(dbname2, options));
}
TEST_F(DBTest2, TraceWithFilter) {
Options options = CurrentOptions();
options.merge_operator = MergeOperators::CreatePutOperator();
ReadOptions ro;
WriteOptions wo;
TraceOptions trace_opts;
EnvOptions env_opts;
CreateAndReopenWithCF({"pikachu"}, options);
Random rnd(301);
Iterator* single_iter = nullptr;
trace_opts.filter = TraceFilterType::kTraceFilterWrite;
std::string trace_filename = dbname_ + "/rocksdb.trace";
std::unique_ptr<TraceWriter> trace_writer;
ASSERT_OK(NewFileTraceWriter(env_, env_opts, trace_filename, &trace_writer));
ASSERT_OK(db_->StartTrace(trace_opts, std::move(trace_writer)));
ASSERT_OK(Put(0, "a", "1"));
ASSERT_OK(Merge(0, "b", "2"));
ASSERT_OK(Delete(0, "c"));
ASSERT_OK(SingleDelete(0, "d"));
ASSERT_OK(db_->DeleteRange(wo, dbfull()->DefaultColumnFamily(), "e", "f"));
WriteBatch batch;
ASSERT_OK(batch.Put("f", "11"));
ASSERT_OK(batch.Merge("g", "12"));
ASSERT_OK(batch.Delete("h"));
ASSERT_OK(batch.SingleDelete("i"));
ASSERT_OK(batch.DeleteRange("j", "k"));
ASSERT_OK(db_->Write(wo, &batch));
single_iter = db_->NewIterator(ro);
single_iter->Seek("f");
single_iter->SeekForPrev("g");
delete single_iter;
ASSERT_EQ("1", Get(0, "a"));
ASSERT_EQ("12", Get(0, "g"));
ASSERT_OK(Put(1, "foo", "bar"));
ASSERT_OK(Put(1, "rocksdb", "rocks"));
ASSERT_EQ("NOT_FOUND", Get(1, "leveldb"));
ASSERT_OK(db_->EndTrace());
// These should not get into the trace file as it is after EndTrace.
ASSERT_OK(Put("hello", "world"));
ASSERT_OK(Merge("foo", "bar"));
// Open another db, replay, and verify the data
std::string value;
std::string dbname2 = test::PerThreadDBPath(env_, "db_replay");
ASSERT_OK(DestroyDB(dbname2, options));
// Using a different name than db2, to pacify infer's use-after-lifetime
// warnings (http://fbinfer.com).
DB* db2_init = nullptr;
options.create_if_missing = true;
ASSERT_OK(DB::Open(options, dbname2, &db2_init));
ColumnFamilyHandle* cf;
ASSERT_OK(
db2_init->CreateColumnFamily(ColumnFamilyOptions(), "pikachu", &cf));
delete cf;
delete db2_init;
DB* db2 = nullptr;
std::vector<ColumnFamilyDescriptor> column_families;
ColumnFamilyOptions cf_options;
cf_options.merge_operator = MergeOperators::CreatePutOperator();
column_families.push_back(ColumnFamilyDescriptor("default", cf_options));
column_families.push_back(
ColumnFamilyDescriptor("pikachu", ColumnFamilyOptions()));
std::vector<ColumnFamilyHandle*> handles;
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
DBOptions db_opts;
db_opts.env = env_;
ASSERT_OK(DB::Open(db_opts, dbname2, column_families, &handles, &db2));
env_->SleepForMicroseconds(100);
// Verify that the keys don't already exist
ASSERT_TRUE(db2->Get(ro, handles[0], "a", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "g", &value).IsNotFound());
std::unique_ptr<TraceReader> trace_reader;
ASSERT_OK(NewFileTraceReader(env_, env_opts, trace_filename, &trace_reader));
std::unique_ptr<Replayer> replayer;
ASSERT_OK(
db2->NewDefaultReplayer(handles, std::move(trace_reader), &replayer));
ASSERT_OK(replayer->Prepare());
ASSERT_OK(replayer->Replay(ReplayOptions(), nullptr));
replayer.reset();
// All the key-values should not present since we filter out the WRITE ops.
ASSERT_TRUE(db2->Get(ro, handles[0], "a", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "g", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "hello", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "world", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "foo", &value).IsNotFound());
ASSERT_TRUE(db2->Get(ro, handles[0], "rocksdb", &value).IsNotFound());
for (auto handle : handles) {
delete handle;
}
delete db2;
ASSERT_OK(DestroyDB(dbname2, options));
// Set up a new db.
std::string dbname3 = test::PerThreadDBPath(env_, "db_not_trace_read");
ASSERT_OK(DestroyDB(dbname3, options));
DB* db3_init = nullptr;
options.create_if_missing = true;
ColumnFamilyHandle* cf3;
ASSERT_OK(DB::Open(options, dbname3, &db3_init));
ASSERT_OK(
db3_init->CreateColumnFamily(ColumnFamilyOptions(), "pikachu", &cf3));
delete cf3;
delete db3_init;
column_families.clear();
column_families.push_back(ColumnFamilyDescriptor("default", cf_options));
column_families.push_back(
ColumnFamilyDescriptor("pikachu", ColumnFamilyOptions()));
handles.clear();
DB* db3 = nullptr;
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
ASSERT_OK(DB::Open(db_opts, dbname3, column_families, &handles, &db3));
env_->SleepForMicroseconds(100);
// Verify that the keys don't already exist
ASSERT_TRUE(db3->Get(ro, handles[0], "a", &value).IsNotFound());
ASSERT_TRUE(db3->Get(ro, handles[0], "g", &value).IsNotFound());
//The tracer will not record the READ ops.
trace_opts.filter = TraceFilterType::kTraceFilterGet;
std::string trace_filename3 = dbname_ + "/rocksdb.trace_3";
std::unique_ptr<TraceWriter> trace_writer3;
ASSERT_OK(
NewFileTraceWriter(env_, env_opts, trace_filename3, &trace_writer3));
ASSERT_OK(db3->StartTrace(trace_opts, std::move(trace_writer3)));
ASSERT_OK(db3->Put(wo, handles[0], "a", "1"));
ASSERT_OK(db3->Merge(wo, handles[0], "b", "2"));
ASSERT_OK(db3->Delete(wo, handles[0], "c"));
ASSERT_OK(db3->SingleDelete(wo, handles[0], "d"));
ASSERT_OK(db3->Get(ro, handles[0], "a", &value));
ASSERT_EQ(value, "1");
ASSERT_TRUE(db3->Get(ro, handles[0], "c", &value).IsNotFound());
ASSERT_OK(db3->EndTrace());
for (auto handle : handles) {
delete handle;
}
delete db3;
ASSERT_OK(DestroyDB(dbname3, options));
std::unique_ptr<TraceReader> trace_reader3;
ASSERT_OK(
NewFileTraceReader(env_, env_opts, trace_filename3, &trace_reader3));
// Count the number of records in the trace file;
int count = 0;
std::string data;
Status s;
while (true) {
s = trace_reader3->Read(&data);
if (!s.ok()) {
break;
}
count += 1;
}
// We also need to count the header and footer
// 4 WRITE + HEADER + FOOTER = 6
ASSERT_EQ(count, 6);
}
#endif // ROCKSDB_LITE
Copy Get() result when file reads use mmap Summary: For iterator reads, a `SuperVersion` is pinned to preserve a snapshot of SST files, and `Block`s are pinned to allow `key()` and `value()` to return pointers directly into a RocksDB memory region. This works for both non-mmap reads, where the block owns the memory region, and mmap reads, where the file owns the memory region. For point reads with `PinnableSlice`, only the `Block` object is pinned. This works for non-mmap reads because the block owns the memory region, so even if the file is deleted after compaction, the memory region survives. However, for mmap reads, file deletion causes the memory region to which the `PinnableSlice` refers to be unmapped. The result is usually a segfault upon accessing the `PinnableSlice`, although sometimes it returned wrong results (I repro'd this a bunch of times with `db_stress`). This PR copies the value into the `PinnableSlice` when it comes from mmap'd memory. We can tell whether the `Block` owns its memory using `Block::cachable()`, which is unset when reads do not use the provided buffer as is the case with mmap file reads. When that is false we ensure the result of `Get()` is copied. This feels like a short-term solution as ideally we'd have the `PinnableSlice` pin the mmap'd memory so we can do zero-copy reads. It seemed hard so I chose this approach to fix correctness in the meantime. Closes https://github.com/facebook/rocksdb/pull/3881 Differential Revision: D8076288 Pulled By: ajkr fbshipit-source-id: 31d78ec010198723522323dbc6ea325122a46b08
7 years ago
TEST_F(DBTest2, PinnableSliceAndMmapReads) {
Options options = CurrentOptions();
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
options.env = env_;
if (!IsMemoryMappedAccessSupported()) {
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
ROCKSDB_GTEST_SKIP("Test requires default environment");
return;
}
Copy Get() result when file reads use mmap Summary: For iterator reads, a `SuperVersion` is pinned to preserve a snapshot of SST files, and `Block`s are pinned to allow `key()` and `value()` to return pointers directly into a RocksDB memory region. This works for both non-mmap reads, where the block owns the memory region, and mmap reads, where the file owns the memory region. For point reads with `PinnableSlice`, only the `Block` object is pinned. This works for non-mmap reads because the block owns the memory region, so even if the file is deleted after compaction, the memory region survives. However, for mmap reads, file deletion causes the memory region to which the `PinnableSlice` refers to be unmapped. The result is usually a segfault upon accessing the `PinnableSlice`, although sometimes it returned wrong results (I repro'd this a bunch of times with `db_stress`). This PR copies the value into the `PinnableSlice` when it comes from mmap'd memory. We can tell whether the `Block` owns its memory using `Block::cachable()`, which is unset when reads do not use the provided buffer as is the case with mmap file reads. When that is false we ensure the result of `Get()` is copied. This feels like a short-term solution as ideally we'd have the `PinnableSlice` pin the mmap'd memory so we can do zero-copy reads. It seemed hard so I chose this approach to fix correctness in the meantime. Closes https://github.com/facebook/rocksdb/pull/3881 Differential Revision: D8076288 Pulled By: ajkr fbshipit-source-id: 31d78ec010198723522323dbc6ea325122a46b08
7 years ago
options.allow_mmap_reads = true;
options.max_open_files = 100;
options.compression = kNoCompression;
Copy Get() result when file reads use mmap Summary: For iterator reads, a `SuperVersion` is pinned to preserve a snapshot of SST files, and `Block`s are pinned to allow `key()` and `value()` to return pointers directly into a RocksDB memory region. This works for both non-mmap reads, where the block owns the memory region, and mmap reads, where the file owns the memory region. For point reads with `PinnableSlice`, only the `Block` object is pinned. This works for non-mmap reads because the block owns the memory region, so even if the file is deleted after compaction, the memory region survives. However, for mmap reads, file deletion causes the memory region to which the `PinnableSlice` refers to be unmapped. The result is usually a segfault upon accessing the `PinnableSlice`, although sometimes it returned wrong results (I repro'd this a bunch of times with `db_stress`). This PR copies the value into the `PinnableSlice` when it comes from mmap'd memory. We can tell whether the `Block` owns its memory using `Block::cachable()`, which is unset when reads do not use the provided buffer as is the case with mmap file reads. When that is false we ensure the result of `Get()` is copied. This feels like a short-term solution as ideally we'd have the `PinnableSlice` pin the mmap'd memory so we can do zero-copy reads. It seemed hard so I chose this approach to fix correctness in the meantime. Closes https://github.com/facebook/rocksdb/pull/3881 Differential Revision: D8076288 Pulled By: ajkr fbshipit-source-id: 31d78ec010198723522323dbc6ea325122a46b08
7 years ago
Reopen(options);
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Flush());
PinnableSlice pinned_value;
ASSERT_EQ(Get("foo", &pinned_value), Status::OK());
// It is not safe to pin mmap files as they might disappear by compaction
ASSERT_FALSE(pinned_value.IsPinned());
Copy Get() result when file reads use mmap Summary: For iterator reads, a `SuperVersion` is pinned to preserve a snapshot of SST files, and `Block`s are pinned to allow `key()` and `value()` to return pointers directly into a RocksDB memory region. This works for both non-mmap reads, where the block owns the memory region, and mmap reads, where the file owns the memory region. For point reads with `PinnableSlice`, only the `Block` object is pinned. This works for non-mmap reads because the block owns the memory region, so even if the file is deleted after compaction, the memory region survives. However, for mmap reads, file deletion causes the memory region to which the `PinnableSlice` refers to be unmapped. The result is usually a segfault upon accessing the `PinnableSlice`, although sometimes it returned wrong results (I repro'd this a bunch of times with `db_stress`). This PR copies the value into the `PinnableSlice` when it comes from mmap'd memory. We can tell whether the `Block` owns its memory using `Block::cachable()`, which is unset when reads do not use the provided buffer as is the case with mmap file reads. When that is false we ensure the result of `Get()` is copied. This feels like a short-term solution as ideally we'd have the `PinnableSlice` pin the mmap'd memory so we can do zero-copy reads. It seemed hard so I chose this approach to fix correctness in the meantime. Closes https://github.com/facebook/rocksdb/pull/3881 Differential Revision: D8076288 Pulled By: ajkr fbshipit-source-id: 31d78ec010198723522323dbc6ea325122a46b08
7 years ago
ASSERT_EQ(pinned_value.ToString(), "bar");
ASSERT_OK(dbfull()->TEST_CompactRange(
0 /* level */, nullptr /* begin */, nullptr /* end */,
nullptr /* column_family */, true /* disallow_trivial_move */));
Copy Get() result when file reads use mmap Summary: For iterator reads, a `SuperVersion` is pinned to preserve a snapshot of SST files, and `Block`s are pinned to allow `key()` and `value()` to return pointers directly into a RocksDB memory region. This works for both non-mmap reads, where the block owns the memory region, and mmap reads, where the file owns the memory region. For point reads with `PinnableSlice`, only the `Block` object is pinned. This works for non-mmap reads because the block owns the memory region, so even if the file is deleted after compaction, the memory region survives. However, for mmap reads, file deletion causes the memory region to which the `PinnableSlice` refers to be unmapped. The result is usually a segfault upon accessing the `PinnableSlice`, although sometimes it returned wrong results (I repro'd this a bunch of times with `db_stress`). This PR copies the value into the `PinnableSlice` when it comes from mmap'd memory. We can tell whether the `Block` owns its memory using `Block::cachable()`, which is unset when reads do not use the provided buffer as is the case with mmap file reads. When that is false we ensure the result of `Get()` is copied. This feels like a short-term solution as ideally we'd have the `PinnableSlice` pin the mmap'd memory so we can do zero-copy reads. It seemed hard so I chose this approach to fix correctness in the meantime. Closes https://github.com/facebook/rocksdb/pull/3881 Differential Revision: D8076288 Pulled By: ajkr fbshipit-source-id: 31d78ec010198723522323dbc6ea325122a46b08
7 years ago
// Ensure pinned_value doesn't rely on memory munmap'd by the above
// compaction. It crashes if it does.
ASSERT_EQ(pinned_value.ToString(), "bar");
#ifndef ROCKSDB_LITE
pinned_value.Reset();
// Unsafe to pin mmap files when they could be kicked out of table cache
Close();
ASSERT_OK(ReadOnlyReopen(options));
ASSERT_EQ(Get("foo", &pinned_value), Status::OK());
ASSERT_FALSE(pinned_value.IsPinned());
ASSERT_EQ(pinned_value.ToString(), "bar");
pinned_value.Reset();
// In read-only mode with infinite capacity on table cache it should pin the
// value and avoid the memcpy
Close();
options.max_open_files = -1;
ASSERT_OK(ReadOnlyReopen(options));
ASSERT_EQ(Get("foo", &pinned_value), Status::OK());
ASSERT_TRUE(pinned_value.IsPinned());
Copy Get() result when file reads use mmap Summary: For iterator reads, a `SuperVersion` is pinned to preserve a snapshot of SST files, and `Block`s are pinned to allow `key()` and `value()` to return pointers directly into a RocksDB memory region. This works for both non-mmap reads, where the block owns the memory region, and mmap reads, where the file owns the memory region. For point reads with `PinnableSlice`, only the `Block` object is pinned. This works for non-mmap reads because the block owns the memory region, so even if the file is deleted after compaction, the memory region survives. However, for mmap reads, file deletion causes the memory region to which the `PinnableSlice` refers to be unmapped. The result is usually a segfault upon accessing the `PinnableSlice`, although sometimes it returned wrong results (I repro'd this a bunch of times with `db_stress`). This PR copies the value into the `PinnableSlice` when it comes from mmap'd memory. We can tell whether the `Block` owns its memory using `Block::cachable()`, which is unset when reads do not use the provided buffer as is the case with mmap file reads. When that is false we ensure the result of `Get()` is copied. This feels like a short-term solution as ideally we'd have the `PinnableSlice` pin the mmap'd memory so we can do zero-copy reads. It seemed hard so I chose this approach to fix correctness in the meantime. Closes https://github.com/facebook/rocksdb/pull/3881 Differential Revision: D8076288 Pulled By: ajkr fbshipit-source-id: 31d78ec010198723522323dbc6ea325122a46b08
7 years ago
ASSERT_EQ(pinned_value.ToString(), "bar");
#endif
Copy Get() result when file reads use mmap Summary: For iterator reads, a `SuperVersion` is pinned to preserve a snapshot of SST files, and `Block`s are pinned to allow `key()` and `value()` to return pointers directly into a RocksDB memory region. This works for both non-mmap reads, where the block owns the memory region, and mmap reads, where the file owns the memory region. For point reads with `PinnableSlice`, only the `Block` object is pinned. This works for non-mmap reads because the block owns the memory region, so even if the file is deleted after compaction, the memory region survives. However, for mmap reads, file deletion causes the memory region to which the `PinnableSlice` refers to be unmapped. The result is usually a segfault upon accessing the `PinnableSlice`, although sometimes it returned wrong results (I repro'd this a bunch of times with `db_stress`). This PR copies the value into the `PinnableSlice` when it comes from mmap'd memory. We can tell whether the `Block` owns its memory using `Block::cachable()`, which is unset when reads do not use the provided buffer as is the case with mmap file reads. When that is false we ensure the result of `Get()` is copied. This feels like a short-term solution as ideally we'd have the `PinnableSlice` pin the mmap'd memory so we can do zero-copy reads. It seemed hard so I chose this approach to fix correctness in the meantime. Closes https://github.com/facebook/rocksdb/pull/3881 Differential Revision: D8076288 Pulled By: ajkr fbshipit-source-id: 31d78ec010198723522323dbc6ea325122a46b08
7 years ago
}
TEST_F(DBTest2, DISABLED_IteratorPinnedMemory) {
Options options = CurrentOptions();
options.create_if_missing = true;
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
BlockBasedTableOptions bbto;
bbto.no_block_cache = false;
bbto.cache_index_and_filter_blocks = false;
bbto.block_cache = NewLRUCache(100000);
bbto.block_size = 400; // small block size
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
options.table_factory.reset(NewBlockBasedTableFactory(bbto));
Reopen(options);
Random rnd(301);
std::string v = rnd.RandomString(400);
// Since v is the size of a block, each key should take a block
// of 400+ bytes.
ASSERT_OK(Put("1", v));
ASSERT_OK(Put("3", v));
ASSERT_OK(Put("5", v));
ASSERT_OK(Put("7", v));
ASSERT_OK(Flush());
ASSERT_EQ(0, bbto.block_cache->GetPinnedUsage());
// Verify that iterators don't pin more than one data block in block cache
// at each time.
{
std::unique_ptr<Iterator> iter(db_->NewIterator(ReadOptions()));
iter->SeekToFirst();
for (int i = 0; i < 4; i++) {
ASSERT_TRUE(iter->Valid());
// Block cache should contain exactly one block.
ASSERT_GT(bbto.block_cache->GetPinnedUsage(), 0);
ASSERT_LT(bbto.block_cache->GetPinnedUsage(), 800);
iter->Next();
}
ASSERT_FALSE(iter->Valid());
iter->Seek("4");
ASSERT_TRUE(iter->Valid());
ASSERT_GT(bbto.block_cache->GetPinnedUsage(), 0);
ASSERT_LT(bbto.block_cache->GetPinnedUsage(), 800);
iter->Seek("3");
ASSERT_TRUE(iter->Valid());
ASSERT_OK(iter->status());
ASSERT_GT(bbto.block_cache->GetPinnedUsage(), 0);
ASSERT_LT(bbto.block_cache->GetPinnedUsage(), 800);
}
ASSERT_EQ(0, bbto.block_cache->GetPinnedUsage());
// Test compaction case
ASSERT_OK(Put("2", v));
ASSERT_OK(Put("5", v));
ASSERT_OK(Put("6", v));
ASSERT_OK(Put("8", v));
ASSERT_OK(Flush());
// Clear existing data in block cache
bbto.block_cache->SetCapacity(0);
bbto.block_cache->SetCapacity(100000);
// Verify compaction input iterators don't hold more than one data blocks at
// one time.
std::atomic<bool> finished(false);
std::atomic<int> block_newed(0);
std::atomic<int> block_destroyed(0);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"Block::Block:0", [&](void* /*arg*/) {
if (finished) {
return;
}
// Two iterators. At most 2 outstanding blocks.
EXPECT_GE(block_newed.load(), block_destroyed.load());
EXPECT_LE(block_newed.load(), block_destroyed.load() + 1);
block_newed.fetch_add(1);
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"Block::~Block", [&](void* /*arg*/) {
if (finished) {
return;
}
// Two iterators. At most 2 outstanding blocks.
EXPECT_GE(block_newed.load(), block_destroyed.load() + 1);
EXPECT_LE(block_newed.load(), block_destroyed.load() + 2);
block_destroyed.fetch_add(1);
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"CompactionJob::Run:BeforeVerify",
[&](void* /*arg*/) { finished = true; });
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
// Two input files. Each of them has 4 data blocks.
ASSERT_EQ(8, block_newed.load());
ASSERT_EQ(8, block_destroyed.load());
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, TestBBTTailPrefetch) {
std::atomic<bool> called(false);
size_t expected_lower_bound = 512 * 1024;
size_t expected_higher_bound = 512 * 1024;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"BlockBasedTable::Open::TailPrefetchLen", [&](void* arg) {
size_t* prefetch_size = static_cast<size_t*>(arg);
EXPECT_LE(expected_lower_bound, *prefetch_size);
EXPECT_GE(expected_higher_bound, *prefetch_size);
called = true;
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
ASSERT_OK(Put("1", "1"));
ASSERT_OK(Put("9", "1"));
ASSERT_OK(Flush());
expected_lower_bound = 0;
expected_higher_bound = 8 * 1024;
ASSERT_OK(Put("1", "1"));
ASSERT_OK(Put("9", "1"));
ASSERT_OK(Flush());
ASSERT_OK(Put("1", "1"));
ASSERT_OK(Put("9", "1"));
ASSERT_OK(Flush());
// Full compaction to make sure there is no L0 file after the open.
ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
ASSERT_TRUE(called.load());
called = false;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearAllCallBacks();
std::atomic<bool> first_call(true);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"BlockBasedTable::Open::TailPrefetchLen", [&](void* arg) {
size_t* prefetch_size = static_cast<size_t*>(arg);
if (first_call) {
EXPECT_EQ(4 * 1024, *prefetch_size);
first_call = false;
} else {
EXPECT_GE(4 * 1024, *prefetch_size);
}
called = true;
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
Options options = CurrentOptions();
options.max_file_opening_threads = 1; // one thread
BlockBasedTableOptions table_options;
table_options.cache_index_and_filter_blocks = true;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.max_open_files = -1;
Reopen(options);
ASSERT_OK(Put("1", "1"));
ASSERT_OK(Put("9", "1"));
ASSERT_OK(Flush());
ASSERT_OK(Put("1", "1"));
ASSERT_OK(Put("9", "1"));
ASSERT_OK(Flush());
ASSERT_TRUE(called.load());
called = false;
// Parallel loading SST files
options.max_file_opening_threads = 16;
Reopen(options);
ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
ASSERT_TRUE(called.load());
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearAllCallBacks();
}
TEST_F(DBTest2, TestGetColumnFamilyHandleUnlocked) {
// Setup sync point dependency to reproduce the race condition of
// DBImpl::GetColumnFamilyHandleUnlocked
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->LoadDependency({
{"TestGetColumnFamilyHandleUnlocked::GetColumnFamilyHandleUnlocked1",
"TestGetColumnFamilyHandleUnlocked::PreGetColumnFamilyHandleUnlocked2"},
{"TestGetColumnFamilyHandleUnlocked::GetColumnFamilyHandleUnlocked2",
"TestGetColumnFamilyHandleUnlocked::ReadColumnFamilyHandle1"},
});
SyncPoint::GetInstance()->EnableProcessing();
CreateColumnFamilies({"test1", "test2"}, Options());
ASSERT_EQ(handles_.size(), 2);
DBImpl* dbi = static_cast_with_check<DBImpl>(db_);
port::Thread user_thread1([&]() {
auto cfh = dbi->GetColumnFamilyHandleUnlocked(handles_[0]->GetID());
ASSERT_EQ(cfh->GetID(), handles_[0]->GetID());
TEST_SYNC_POINT("TestGetColumnFamilyHandleUnlocked::GetColumnFamilyHandleUnlocked1");
TEST_SYNC_POINT("TestGetColumnFamilyHandleUnlocked::ReadColumnFamilyHandle1");
ASSERT_EQ(cfh->GetID(), handles_[0]->GetID());
});
port::Thread user_thread2([&]() {
TEST_SYNC_POINT("TestGetColumnFamilyHandleUnlocked::PreGetColumnFamilyHandleUnlocked2");
auto cfh = dbi->GetColumnFamilyHandleUnlocked(handles_[1]->GetID());
ASSERT_EQ(cfh->GetID(), handles_[1]->GetID());
TEST_SYNC_POINT("TestGetColumnFamilyHandleUnlocked::GetColumnFamilyHandleUnlocked2");
ASSERT_EQ(cfh->GetID(), handles_[1]->GetID());
});
user_thread1.join();
user_thread2.join();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearAllCallBacks();
}
#ifndef ROCKSDB_LITE
TEST_F(DBTest2, TestCompactFiles) {
// Setup sync point dependency to reproduce the race condition of
// DBImpl::GetColumnFamilyHandleUnlocked
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->LoadDependency({
{"TestCompactFiles::IngestExternalFile1",
"TestCompactFiles::IngestExternalFile2"},
});
SyncPoint::GetInstance()->EnableProcessing();
Options options;
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
options.env = env_;
options.num_levels = 2;
options.disable_auto_compactions = true;
Reopen(options);
auto* handle = db_->DefaultColumnFamily();
ASSERT_EQ(db_->NumberLevels(handle), 2);
ROCKSDB_NAMESPACE::SstFileWriter sst_file_writer{
ROCKSDB_NAMESPACE::EnvOptions(), options};
std::string external_file1 = dbname_ + "/test_compact_files1.sst_t";
std::string external_file2 = dbname_ + "/test_compact_files2.sst_t";
std::string external_file3 = dbname_ + "/test_compact_files3.sst_t";
ASSERT_OK(sst_file_writer.Open(external_file1));
ASSERT_OK(sst_file_writer.Put("1", "1"));
ASSERT_OK(sst_file_writer.Put("2", "2"));
ASSERT_OK(sst_file_writer.Finish());
ASSERT_OK(sst_file_writer.Open(external_file2));
ASSERT_OK(sst_file_writer.Put("3", "3"));
ASSERT_OK(sst_file_writer.Put("4", "4"));
ASSERT_OK(sst_file_writer.Finish());
ASSERT_OK(sst_file_writer.Open(external_file3));
ASSERT_OK(sst_file_writer.Put("5", "5"));
ASSERT_OK(sst_file_writer.Put("6", "6"));
ASSERT_OK(sst_file_writer.Finish());
ASSERT_OK(db_->IngestExternalFile(handle, {external_file1, external_file3},
IngestExternalFileOptions()));
ASSERT_EQ(NumTableFilesAtLevel(1, 0), 2);
std::vector<std::string> files;
GetSstFiles(env_, dbname_, &files);
ASSERT_EQ(files.size(), 2);
Status user_thread1_status;
port::Thread user_thread1([&]() {
user_thread1_status =
db_->CompactFiles(CompactionOptions(), handle, files, 1);
});
Status user_thread2_status;
port::Thread user_thread2([&]() {
user_thread2_status = db_->IngestExternalFile(handle, {external_file2},
IngestExternalFileOptions());
TEST_SYNC_POINT("TestCompactFiles::IngestExternalFile1");
});
user_thread1.join();
user_thread2.join();
ASSERT_OK(user_thread1_status);
ASSERT_OK(user_thread2_status);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearAllCallBacks();
}
#endif // ROCKSDB_LITE
TEST_F(DBTest2, MultiDBParallelOpenTest) {
const int kNumDbs = 2;
Options options = CurrentOptions();
std::vector<std::string> dbnames;
for (int i = 0; i < kNumDbs; ++i) {
dbnames.emplace_back(test::PerThreadDBPath(env_, "db" + std::to_string(i)));
ASSERT_OK(DestroyDB(dbnames.back(), options));
}
// Verify empty DBs can be created in parallel
std::vector<std::thread> open_threads;
std::vector<DB*> dbs{static_cast<unsigned int>(kNumDbs), nullptr};
options.create_if_missing = true;
for (int i = 0; i < kNumDbs; ++i) {
open_threads.emplace_back(
[&](int dbnum) {
ASSERT_OK(DB::Open(options, dbnames[dbnum], &dbs[dbnum]));
},
i);
}
// Now add some data and close, so next we can verify non-empty DBs can be
// recovered in parallel
for (int i = 0; i < kNumDbs; ++i) {
open_threads[i].join();
ASSERT_OK(dbs[i]->Put(WriteOptions(), "xi", "gua"));
delete dbs[i];
}
// Verify non-empty DBs can be recovered in parallel
open_threads.clear();
for (int i = 0; i < kNumDbs; ++i) {
open_threads.emplace_back(
[&](int dbnum) {
ASSERT_OK(DB::Open(options, dbnames[dbnum], &dbs[dbnum]));
},
i);
}
// Wait and cleanup
for (int i = 0; i < kNumDbs; ++i) {
open_threads[i].join();
delete dbs[i];
ASSERT_OK(DestroyDB(dbnames[i], options));
}
}
namespace {
class DummyOldStats : public Statistics {
public:
const char* Name() const override { return "DummyOldStats"; }
uint64_t getTickerCount(uint32_t /*ticker_type*/) const override { return 0; }
void recordTick(uint32_t /* ticker_type */, uint64_t /* count */) override {
num_rt++;
}
void setTickerCount(uint32_t /*ticker_type*/, uint64_t /*count*/) override {}
uint64_t getAndResetTickerCount(uint32_t /*ticker_type*/) override {
return 0;
}
void measureTime(uint32_t /*histogram_type*/, uint64_t /*count*/) override {
num_mt++;
}
void histogramData(
uint32_t /*histogram_type*/,
ROCKSDB_NAMESPACE::HistogramData* const /*data*/) const override {}
std::string getHistogramString(uint32_t /*type*/) const override {
return "";
}
bool HistEnabledForType(uint32_t /*type*/) const override { return false; }
std::string ToString() const override { return ""; }
std::atomic<int> num_rt{0};
std::atomic<int> num_mt{0};
};
} // namespace
TEST_F(DBTest2, OldStatsInterface) {
DummyOldStats* dos = new DummyOldStats();
std::shared_ptr<Statistics> stats(dos);
Options options = CurrentOptions();
options.create_if_missing = true;
options.statistics = stats;
Reopen(options);
ASSERT_OK(Put("foo", "bar"));
ASSERT_EQ("bar", Get("foo"));
ASSERT_OK(Flush());
ASSERT_EQ("bar", Get("foo"));
ASSERT_GT(dos->num_rt, 0);
ASSERT_GT(dos->num_mt, 0);
}
TEST_F(DBTest2, CloseWithUnreleasedSnapshot) {
const Snapshot* ss = db_->GetSnapshot();
for (auto h : handles_) {
db_->DestroyColumnFamilyHandle(h);
}
handles_.clear();
ASSERT_NOK(db_->Close());
db_->ReleaseSnapshot(ss);
ASSERT_OK(db_->Close());
delete db_;
db_ = nullptr;
}
TEST_F(DBTest2, PrefixBloomReseek) {
Options options = CurrentOptions();
options.create_if_missing = true;
options.prefix_extractor.reset(NewCappedPrefixTransform(3));
BlockBasedTableOptions bbto;
bbto.filter_policy.reset(NewBloomFilterPolicy(10, false));
bbto.whole_key_filtering = false;
options.table_factory.reset(NewBlockBasedTableFactory(bbto));
DestroyAndReopen(options);
// Construct two L1 files with keys:
// f1:[aaa1 ccc1] f2:[ddd0]
ASSERT_OK(Put("aaa1", ""));
ASSERT_OK(Put("ccc1", ""));
ASSERT_OK(Flush());
ASSERT_OK(Put("ddd0", ""));
ASSERT_OK(Flush());
CompactRangeOptions cro;
cro.bottommost_level_compaction = BottommostLevelCompaction::kSkip;
ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
ASSERT_OK(Put("bbb1", ""));
Iterator* iter = db_->NewIterator(ReadOptions());
ASSERT_OK(iter->status());
// Seeking into f1, the iterator will check bloom filter which returns the
// file iterator ot be invalidate, and the cursor will put into f2, with
// the next key to be "ddd0".
iter->Seek("bbb1");
ASSERT_TRUE(iter->Valid());
ASSERT_EQ("bbb1", iter->key().ToString());
// Reseek ccc1, the L1 iterator needs to go back to f1 and reseek.
iter->Seek("ccc1");
ASSERT_TRUE(iter->Valid());
ASSERT_EQ("ccc1", iter->key().ToString());
delete iter;
}
TEST_F(DBTest2, PrefixBloomFilteredOut) {
Options options = CurrentOptions();
options.create_if_missing = true;
options.prefix_extractor.reset(NewCappedPrefixTransform(3));
BlockBasedTableOptions bbto;
bbto.filter_policy.reset(NewBloomFilterPolicy(10, false));
bbto.whole_key_filtering = false;
options.table_factory.reset(NewBlockBasedTableFactory(bbto));
DestroyAndReopen(options);
// Construct two L1 files with keys:
// f1:[aaa1 ccc1] f2:[ddd0]
ASSERT_OK(Put("aaa1", ""));
ASSERT_OK(Put("ccc1", ""));
ASSERT_OK(Flush());
ASSERT_OK(Put("ddd0", ""));
ASSERT_OK(Flush());
CompactRangeOptions cro;
cro.bottommost_level_compaction = BottommostLevelCompaction::kSkip;
ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
Iterator* iter = db_->NewIterator(ReadOptions());
ASSERT_OK(iter->status());
// Bloom filter is filterd out by f1.
// This is just one of several valid position following the contract.
// Postioning to ccc1 or ddd0 is also valid. This is just to validate
// the behavior of the current implementation. If underlying implementation
// changes, the test might fail here.
iter->Seek("bbb1");
ASSERT_OK(iter->status());
ASSERT_FALSE(iter->Valid());
delete iter;
}
#ifndef ROCKSDB_LITE
TEST_F(DBTest2, RowCacheSnapshot) {
Options options = CurrentOptions();
options.statistics = ROCKSDB_NAMESPACE::CreateDBStatistics();
options.row_cache = NewLRUCache(8 * 8192);
DestroyAndReopen(options);
ASSERT_OK(Put("foo", "bar1"));
const Snapshot* s1 = db_->GetSnapshot();
ASSERT_OK(Put("foo", "bar2"));
ASSERT_OK(Flush());
ASSERT_OK(Put("foo2", "bar"));
const Snapshot* s2 = db_->GetSnapshot();
ASSERT_OK(Put("foo3", "bar"));
const Snapshot* s3 = db_->GetSnapshot();
ASSERT_EQ(TestGetTickerCount(options, ROW_CACHE_HIT), 0);
ASSERT_EQ(TestGetTickerCount(options, ROW_CACHE_MISS), 0);
ASSERT_EQ(Get("foo"), "bar2");
ASSERT_EQ(TestGetTickerCount(options, ROW_CACHE_HIT), 0);
ASSERT_EQ(TestGetTickerCount(options, ROW_CACHE_MISS), 1);
ASSERT_EQ(Get("foo"), "bar2");
ASSERT_EQ(TestGetTickerCount(options, ROW_CACHE_HIT), 1);
ASSERT_EQ(TestGetTickerCount(options, ROW_CACHE_MISS), 1);
ASSERT_EQ(Get("foo", s1), "bar1");
ASSERT_EQ(TestGetTickerCount(options, ROW_CACHE_HIT), 1);
ASSERT_EQ(TestGetTickerCount(options, ROW_CACHE_MISS), 2);
ASSERT_EQ(Get("foo", s2), "bar2");
ASSERT_EQ(TestGetTickerCount(options, ROW_CACHE_HIT), 2);
ASSERT_EQ(TestGetTickerCount(options, ROW_CACHE_MISS), 2);
ASSERT_EQ(Get("foo", s1), "bar1");
ASSERT_EQ(TestGetTickerCount(options, ROW_CACHE_HIT), 3);
ASSERT_EQ(TestGetTickerCount(options, ROW_CACHE_MISS), 2);
ASSERT_EQ(Get("foo", s3), "bar2");
ASSERT_EQ(TestGetTickerCount(options, ROW_CACHE_HIT), 4);
ASSERT_EQ(TestGetTickerCount(options, ROW_CACHE_MISS), 2);
db_->ReleaseSnapshot(s1);
db_->ReleaseSnapshot(s2);
db_->ReleaseSnapshot(s3);
}
#endif // ROCKSDB_LITE
// When DB is reopened with multiple column families, the manifest file
// is written after the first CF is flushed, and it is written again
// after each flush. If DB crashes between the flushes, the flushed CF
// flushed will pass the latest log file, and now we require it not
// to be corrupted, and triggering a corruption report.
// We need to fix the bug and enable the test.
TEST_F(DBTest2, CrashInRecoveryMultipleCF) {
const std::vector<std::string> sync_points = {
"DBImpl::RecoverLogFiles:BeforeFlushFinalMemtable",
"VersionSet::ProcessManifestWrites:BeforeWriteLastVersionEdit:0"};
for (const auto& test_sync_point : sync_points) {
Options options = CurrentOptions();
// First destroy original db to ensure a clean start.
DestroyAndReopen(options);
options.create_if_missing = true;
options.wal_recovery_mode = WALRecoveryMode::kPointInTimeRecovery;
CreateAndReopenWithCF({"pikachu"}, options);
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Flush());
ASSERT_OK(Put(1, "foo", "bar"));
ASSERT_OK(Flush(1));
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Put(1, "foo", "bar"));
// The value is large enough to be divided to two blocks.
std::string large_value(400, ' ');
ASSERT_OK(Put("foo1", large_value));
ASSERT_OK(Put("foo2", large_value));
Close();
// Corrupt the log file in the middle, so that it is not corrupted
// in the tail.
std::vector<std::string> filenames;
ASSERT_OK(env_->GetChildren(dbname_, &filenames));
for (const auto& f : filenames) {
uint64_t number;
FileType type;
if (ParseFileName(f, &number, &type) && type == FileType::kWalFile) {
std::string fname = dbname_ + "/" + f;
std::string file_content;
ASSERT_OK(ReadFileToString(env_, fname, &file_content));
file_content[400] = 'h';
file_content[401] = 'a';
ASSERT_OK(WriteStringToFile(env_, file_content, fname));
break;
}
}
// Reopen and freeze the file system after the first manifest write.
FaultInjectionTestEnv fit_env(options.env);
options.env = &fit_env;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->ClearAllCallBacks();
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
test_sync_point,
[&](void* /*arg*/) { fit_env.SetFilesystemActive(false); });
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
ASSERT_NOK(TryReopenWithColumnFamilies(
{kDefaultColumnFamilyName, "pikachu"}, options));
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
fit_env.SetFilesystemActive(true);
// If we continue using failure ingestion Env, it will conplain something
// when renaming current file, which is not expected. Need to investigate
// why.
options.env = env_;
ASSERT_OK(TryReopenWithColumnFamilies({kDefaultColumnFamilyName, "pikachu"},
options));
}
}
TEST_F(DBTest2, SeekFileRangeDeleteTail) {
Options options = CurrentOptions();
options.prefix_extractor.reset(NewCappedPrefixTransform(1));
options.num_levels = 3;
DestroyAndReopen(options);
ASSERT_OK(Put("a", "a"));
const Snapshot* s1 = db_->GetSnapshot();
ASSERT_OK(
db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "a", "f"));
ASSERT_OK(Put("b", "a"));
ASSERT_OK(Flush());
ASSERT_OK(Put("x", "a"));
ASSERT_OK(Put("z", "a"));
ASSERT_OK(Flush());
CompactRangeOptions cro;
cro.change_level = true;
cro.target_level = 2;
ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
{
ReadOptions ro;
ro.total_order_seek = true;
std::unique_ptr<Iterator> iter(db_->NewIterator(ro));
ASSERT_OK(iter->status());
iter->Seek("e");
ASSERT_TRUE(iter->Valid());
ASSERT_EQ("x", iter->key().ToString());
}
db_->ReleaseSnapshot(s1);
}
TEST_F(DBTest2, BackgroundPurgeTest) {
Options options = CurrentOptions();
options.write_buffer_manager =
std::make_shared<ROCKSDB_NAMESPACE::WriteBufferManager>(1 << 20);
options.avoid_unnecessary_blocking_io = true;
DestroyAndReopen(options);
size_t base_value = options.write_buffer_manager->memory_usage();
ASSERT_OK(Put("a", "a"));
Iterator* iter = db_->NewIterator(ReadOptions());
ASSERT_OK(iter->status());
ASSERT_OK(Flush());
size_t value = options.write_buffer_manager->memory_usage();
ASSERT_GT(value, base_value);
db_->GetEnv()->SetBackgroundThreads(1, Env::Priority::HIGH);
test::SleepingBackgroundTask sleeping_task_after;
db_->GetEnv()->Schedule(&test::SleepingBackgroundTask::DoSleepTask,
&sleeping_task_after, Env::Priority::HIGH);
delete iter;
Env::Default()->SleepForMicroseconds(100000);
value = options.write_buffer_manager->memory_usage();
ASSERT_GT(value, base_value);
sleeping_task_after.WakeUp();
sleeping_task_after.WaitUntilDone();
test::SleepingBackgroundTask sleeping_task_after2;
db_->GetEnv()->Schedule(&test::SleepingBackgroundTask::DoSleepTask,
&sleeping_task_after2, Env::Priority::HIGH);
sleeping_task_after2.WakeUp();
sleeping_task_after2.WaitUntilDone();
value = options.write_buffer_manager->memory_usage();
ASSERT_EQ(base_value, value);
}
Fix a data race for cfd->log_number_ (#6249) Summary: A thread calling LogAndApply may release db mutex when calling WriteCurrentStateToManifest() which reads cfd->log_number_. Another thread can call SwitchMemtable() and writes to cfd->log_number_. Solution is to cache the cfd->log_number_ before releasing mutex in LogAndApply. Test Plan (on devserver): ``` $COMPILE_WITH_TSAN=1 make db_stress $./db_stress --acquire_snapshot_one_in=10000 --avoid_unnecessary_blocking_io=1 --block_size=16384 --bloom_bits=16 --bottommost_compression_type=zstd --cache_index_and_filter_blocks=1 --cache_size=1048576 --checkpoint_one_in=1000000 --checksum_type=kxxHash --clear_column_family_one_in=0 --compact_files_one_in=1000000 --compact_range_one_in=1000000 --compaction_ttl=0 --compression_max_dict_bytes=16384 --compression_type=zstd --compression_zstd_max_train_bytes=0 --continuous_verification_interval=0 --db=/dev/shm/rocksdb/rocksdb_crashtest_blackbox --db_write_buffer_size=1048576 --delpercent=5 --delrangepercent=0 --destroy_db_initially=0 --enable_pipelined_write=0 --flush_one_in=1000000 --format_version=5 --get_live_files_and_wal_files_one_in=1000000 --index_block_restart_interval=5 --index_type=0 --log2_keys_per_lock=22 --long_running_snapshots=0 --max_background_compactions=20 --max_bytes_for_level_base=10485760 --max_key=1000000 --max_manifest_file_size=16384 --max_write_batch_group_size_bytes=16 --max_write_buffer_number=3 --memtablerep=skip_list --mmap_read=0 --nooverwritepercent=1 --open_files=500000 --ops_per_thread=100000000 --partition_filters=0 --pause_background_one_in=1000000 --periodic_compaction_seconds=0 --prefixpercent=5 --progress_reports=0 --readpercent=45 --recycle_log_file_num=0 --reopen=20 --set_options_one_in=10000 --snapshot_hold_ops=100000 --subcompactions=2 --sync=1 --target_file_size_base=2097152 --target_file_size_multiplier=2 --test_batches_snapshots=1 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --use_full_merge_v1=0 --use_merge=0 --use_multiget=1 --verify_checksum=1 --verify_checksum_one_in=1000000 --verify_db_one_in=100000 --write_buffer_size=4194304 --write_dbid_to_manifest=1 --writepercent=35 ``` Then repeat the following multiple times, e.g. 100 after compiling with tsan. ``` $./db_test2 --gtest_filter=DBTest2.SwitchMemtableRaceWithNewManifest ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6249 Differential Revision: D19235077 Pulled By: riversand963 fbshipit-source-id: 79467b52f48739ce7c27e440caa2447a40653173
5 years ago
TEST_F(DBTest2, SwitchMemtableRaceWithNewManifest) {
Options options = CurrentOptions();
DestroyAndReopen(options);
options.max_manifest_file_size = 10;
options.create_if_missing = true;
CreateAndReopenWithCF({"pikachu"}, options);
ASSERT_EQ(2, handles_.size());
Fix a data race for cfd->log_number_ (#6249) Summary: A thread calling LogAndApply may release db mutex when calling WriteCurrentStateToManifest() which reads cfd->log_number_. Another thread can call SwitchMemtable() and writes to cfd->log_number_. Solution is to cache the cfd->log_number_ before releasing mutex in LogAndApply. Test Plan (on devserver): ``` $COMPILE_WITH_TSAN=1 make db_stress $./db_stress --acquire_snapshot_one_in=10000 --avoid_unnecessary_blocking_io=1 --block_size=16384 --bloom_bits=16 --bottommost_compression_type=zstd --cache_index_and_filter_blocks=1 --cache_size=1048576 --checkpoint_one_in=1000000 --checksum_type=kxxHash --clear_column_family_one_in=0 --compact_files_one_in=1000000 --compact_range_one_in=1000000 --compaction_ttl=0 --compression_max_dict_bytes=16384 --compression_type=zstd --compression_zstd_max_train_bytes=0 --continuous_verification_interval=0 --db=/dev/shm/rocksdb/rocksdb_crashtest_blackbox --db_write_buffer_size=1048576 --delpercent=5 --delrangepercent=0 --destroy_db_initially=0 --enable_pipelined_write=0 --flush_one_in=1000000 --format_version=5 --get_live_files_and_wal_files_one_in=1000000 --index_block_restart_interval=5 --index_type=0 --log2_keys_per_lock=22 --long_running_snapshots=0 --max_background_compactions=20 --max_bytes_for_level_base=10485760 --max_key=1000000 --max_manifest_file_size=16384 --max_write_batch_group_size_bytes=16 --max_write_buffer_number=3 --memtablerep=skip_list --mmap_read=0 --nooverwritepercent=1 --open_files=500000 --ops_per_thread=100000000 --partition_filters=0 --pause_background_one_in=1000000 --periodic_compaction_seconds=0 --prefixpercent=5 --progress_reports=0 --readpercent=45 --recycle_log_file_num=0 --reopen=20 --set_options_one_in=10000 --snapshot_hold_ops=100000 --subcompactions=2 --sync=1 --target_file_size_base=2097152 --target_file_size_multiplier=2 --test_batches_snapshots=1 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --use_full_merge_v1=0 --use_merge=0 --use_multiget=1 --verify_checksum=1 --verify_checksum_one_in=1000000 --verify_db_one_in=100000 --write_buffer_size=4194304 --write_dbid_to_manifest=1 --writepercent=35 ``` Then repeat the following multiple times, e.g. 100 after compiling with tsan. ``` $./db_test2 --gtest_filter=DBTest2.SwitchMemtableRaceWithNewManifest ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6249 Differential Revision: D19235077 Pulled By: riversand963 fbshipit-source-id: 79467b52f48739ce7c27e440caa2447a40653173
5 years ago
ASSERT_OK(Put("foo", "value"));
const int kL0Files = options.level0_file_num_compaction_trigger;
for (int i = 0; i < kL0Files; ++i) {
ASSERT_OK(Put(/*cf=*/1, "a", std::to_string(i)));
ASSERT_OK(Flush(/*cf=*/1));
}
Fix a data race for cfd->log_number_ (#6249) Summary: A thread calling LogAndApply may release db mutex when calling WriteCurrentStateToManifest() which reads cfd->log_number_. Another thread can call SwitchMemtable() and writes to cfd->log_number_. Solution is to cache the cfd->log_number_ before releasing mutex in LogAndApply. Test Plan (on devserver): ``` $COMPILE_WITH_TSAN=1 make db_stress $./db_stress --acquire_snapshot_one_in=10000 --avoid_unnecessary_blocking_io=1 --block_size=16384 --bloom_bits=16 --bottommost_compression_type=zstd --cache_index_and_filter_blocks=1 --cache_size=1048576 --checkpoint_one_in=1000000 --checksum_type=kxxHash --clear_column_family_one_in=0 --compact_files_one_in=1000000 --compact_range_one_in=1000000 --compaction_ttl=0 --compression_max_dict_bytes=16384 --compression_type=zstd --compression_zstd_max_train_bytes=0 --continuous_verification_interval=0 --db=/dev/shm/rocksdb/rocksdb_crashtest_blackbox --db_write_buffer_size=1048576 --delpercent=5 --delrangepercent=0 --destroy_db_initially=0 --enable_pipelined_write=0 --flush_one_in=1000000 --format_version=5 --get_live_files_and_wal_files_one_in=1000000 --index_block_restart_interval=5 --index_type=0 --log2_keys_per_lock=22 --long_running_snapshots=0 --max_background_compactions=20 --max_bytes_for_level_base=10485760 --max_key=1000000 --max_manifest_file_size=16384 --max_write_batch_group_size_bytes=16 --max_write_buffer_number=3 --memtablerep=skip_list --mmap_read=0 --nooverwritepercent=1 --open_files=500000 --ops_per_thread=100000000 --partition_filters=0 --pause_background_one_in=1000000 --periodic_compaction_seconds=0 --prefixpercent=5 --progress_reports=0 --readpercent=45 --recycle_log_file_num=0 --reopen=20 --set_options_one_in=10000 --snapshot_hold_ops=100000 --subcompactions=2 --sync=1 --target_file_size_base=2097152 --target_file_size_multiplier=2 --test_batches_snapshots=1 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --use_full_merge_v1=0 --use_merge=0 --use_multiget=1 --verify_checksum=1 --verify_checksum_one_in=1000000 --verify_db_one_in=100000 --write_buffer_size=4194304 --write_dbid_to_manifest=1 --writepercent=35 ``` Then repeat the following multiple times, e.g. 100 after compiling with tsan. ``` $./db_test2 --gtest_filter=DBTest2.SwitchMemtableRaceWithNewManifest ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6249 Differential Revision: D19235077 Pulled By: riversand963 fbshipit-source-id: 79467b52f48739ce7c27e440caa2447a40653173
5 years ago
port::Thread thread([&]() { ASSERT_OK(Flush()); });
ASSERT_OK(dbfull()->TEST_WaitForCompact());
Fix a data race for cfd->log_number_ (#6249) Summary: A thread calling LogAndApply may release db mutex when calling WriteCurrentStateToManifest() which reads cfd->log_number_. Another thread can call SwitchMemtable() and writes to cfd->log_number_. Solution is to cache the cfd->log_number_ before releasing mutex in LogAndApply. Test Plan (on devserver): ``` $COMPILE_WITH_TSAN=1 make db_stress $./db_stress --acquire_snapshot_one_in=10000 --avoid_unnecessary_blocking_io=1 --block_size=16384 --bloom_bits=16 --bottommost_compression_type=zstd --cache_index_and_filter_blocks=1 --cache_size=1048576 --checkpoint_one_in=1000000 --checksum_type=kxxHash --clear_column_family_one_in=0 --compact_files_one_in=1000000 --compact_range_one_in=1000000 --compaction_ttl=0 --compression_max_dict_bytes=16384 --compression_type=zstd --compression_zstd_max_train_bytes=0 --continuous_verification_interval=0 --db=/dev/shm/rocksdb/rocksdb_crashtest_blackbox --db_write_buffer_size=1048576 --delpercent=5 --delrangepercent=0 --destroy_db_initially=0 --enable_pipelined_write=0 --flush_one_in=1000000 --format_version=5 --get_live_files_and_wal_files_one_in=1000000 --index_block_restart_interval=5 --index_type=0 --log2_keys_per_lock=22 --long_running_snapshots=0 --max_background_compactions=20 --max_bytes_for_level_base=10485760 --max_key=1000000 --max_manifest_file_size=16384 --max_write_batch_group_size_bytes=16 --max_write_buffer_number=3 --memtablerep=skip_list --mmap_read=0 --nooverwritepercent=1 --open_files=500000 --ops_per_thread=100000000 --partition_filters=0 --pause_background_one_in=1000000 --periodic_compaction_seconds=0 --prefixpercent=5 --progress_reports=0 --readpercent=45 --recycle_log_file_num=0 --reopen=20 --set_options_one_in=10000 --snapshot_hold_ops=100000 --subcompactions=2 --sync=1 --target_file_size_base=2097152 --target_file_size_multiplier=2 --test_batches_snapshots=1 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --use_full_merge_v1=0 --use_merge=0 --use_multiget=1 --verify_checksum=1 --verify_checksum_one_in=1000000 --verify_db_one_in=100000 --write_buffer_size=4194304 --write_dbid_to_manifest=1 --writepercent=35 ``` Then repeat the following multiple times, e.g. 100 after compiling with tsan. ``` $./db_test2 --gtest_filter=DBTest2.SwitchMemtableRaceWithNewManifest ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/6249 Differential Revision: D19235077 Pulled By: riversand963 fbshipit-source-id: 79467b52f48739ce7c27e440caa2447a40653173
5 years ago
thread.join();
}
TEST_F(DBTest2, SameSmallestInSameLevel) {
// This test validates fractional casacading logic when several files at one
// one level only contains the same user key.
Options options = CurrentOptions();
options.merge_operator = MergeOperators::CreateStringAppendOperator();
DestroyAndReopen(options);
ASSERT_OK(Put("key", "1"));
ASSERT_OK(Put("key", "2"));
ASSERT_OK(db_->Merge(WriteOptions(), "key", "3"));
ASSERT_OK(db_->Merge(WriteOptions(), "key", "4"));
ASSERT_OK(Flush());
CompactRangeOptions cro;
cro.change_level = true;
cro.target_level = 2;
ASSERT_OK(dbfull()->CompactRange(cro, db_->DefaultColumnFamily(), nullptr,
nullptr));
ASSERT_OK(db_->Merge(WriteOptions(), "key", "5"));
ASSERT_OK(Flush());
ASSERT_OK(db_->Merge(WriteOptions(), "key", "6"));
ASSERT_OK(Flush());
ASSERT_OK(db_->Merge(WriteOptions(), "key", "7"));
ASSERT_OK(Flush());
ASSERT_OK(db_->Merge(WriteOptions(), "key", "8"));
ASSERT_OK(Flush());
ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
#ifndef ROCKSDB_LITE
ASSERT_EQ("0,4,1", FilesPerLevel());
#endif // ROCKSDB_LITE
ASSERT_EQ("2,3,4,5,6,7,8", Get("key"));
}
TEST_F(DBTest2, FileConsistencyCheckInOpen) {
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Flush());
SyncPoint::GetInstance()->SetCallBack(
"VersionBuilder::CheckConsistencyBeforeReturn", [&](void* arg) {
Status* ret_s = static_cast<Status*>(arg);
*ret_s = Status::Corruption("fcc");
});
SyncPoint::GetInstance()->EnableProcessing();
Options options = CurrentOptions();
options.force_consistency_checks = true;
ASSERT_NOK(TryReopen(options));
SyncPoint::GetInstance()->DisableProcessing();
}
TEST_F(DBTest2, BlockBasedTablePrefixIndexSeekForPrev) {
// create a DB with block prefix index
BlockBasedTableOptions table_options;
Options options = CurrentOptions();
table_options.block_size = 300;
table_options.index_type = BlockBasedTableOptions::kHashSearch;
table_options.index_shortening =
BlockBasedTableOptions::IndexShorteningMode::kNoShortening;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.prefix_extractor.reset(NewFixedPrefixTransform(1));
Reopen(options);
Random rnd(301);
std::string large_value = rnd.RandomString(500);
ASSERT_OK(Put("a1", large_value));
ASSERT_OK(Put("x1", large_value));
ASSERT_OK(Put("y1", large_value));
ASSERT_OK(Flush());
{
std::unique_ptr<Iterator> iterator(db_->NewIterator(ReadOptions()));
ASSERT_OK(iterator->status());
iterator->SeekForPrev("x3");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("x1", iterator->key().ToString());
iterator->SeekForPrev("a3");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("a1", iterator->key().ToString());
iterator->SeekForPrev("y3");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("y1", iterator->key().ToString());
// Query more than one non-existing prefix to cover the case both
// of empty hash bucket and hash bucket conflict.
iterator->SeekForPrev("b1");
// Result should be not valid or "a1".
if (iterator->Valid()) {
ASSERT_EQ("a1", iterator->key().ToString());
}
iterator->SeekForPrev("c1");
// Result should be not valid or "a1".
if (iterator->Valid()) {
ASSERT_EQ("a1", iterator->key().ToString());
}
iterator->SeekForPrev("d1");
// Result should be not valid or "a1".
if (iterator->Valid()) {
ASSERT_EQ("a1", iterator->key().ToString());
}
iterator->SeekForPrev("y3");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("y1", iterator->key().ToString());
}
}
TEST_F(DBTest2, PartitionedIndexPrefetchFailure) {
Options options = last_options_;
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
options.env = env_;
options.max_open_files = 20;
BlockBasedTableOptions bbto;
bbto.index_type = BlockBasedTableOptions::IndexType::kTwoLevelIndexSearch;
bbto.metadata_block_size = 128;
bbto.block_size = 128;
bbto.block_cache = NewLRUCache(16777216);
bbto.cache_index_and_filter_blocks = true;
options.table_factory.reset(NewBlockBasedTableFactory(bbto));
DestroyAndReopen(options);
// Force no table cache so every read will preload the SST file.
dbfull()->TEST_table_cache()->SetCapacity(0);
bbto.block_cache->SetCapacity(0);
Random rnd(301);
for (int i = 0; i < 4096; i++) {
ASSERT_OK(Put(Key(i), rnd.RandomString(32)));
}
ASSERT_OK(Flush());
// Try different random failures in table open for 300 times.
for (int i = 0; i < 300; i++) {
env_->num_reads_fails_ = 0;
env_->rand_reads_fail_odd_ = 8;
std::string value;
Status s = dbfull()->Get(ReadOptions(), Key(1), &value);
if (env_->num_reads_fails_ > 0) {
ASSERT_NOK(s);
} else {
ASSERT_OK(s);
}
}
env_->rand_reads_fail_odd_ = 0;
}
TEST_F(DBTest2, ChangePrefixExtractor) {
for (bool use_partitioned_filter : {true, false}) {
// create a DB with block prefix index
BlockBasedTableOptions table_options;
Options options = CurrentOptions();
// Sometimes filter is checked based on upper bound. Assert counters
// for that case. Otherwise, only check data correctness.
#ifndef ROCKSDB_LITE
bool expect_filter_check = !use_partitioned_filter;
#else
bool expect_filter_check = false;
#endif
table_options.partition_filters = use_partitioned_filter;
if (use_partitioned_filter) {
table_options.index_type =
BlockBasedTableOptions::IndexType::kTwoLevelIndexSearch;
}
table_options.filter_policy.reset(NewBloomFilterPolicy(10, false));
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.statistics = CreateDBStatistics();
options.prefix_extractor.reset(NewFixedPrefixTransform(2));
DestroyAndReopen(options);
Random rnd(301);
ASSERT_OK(Put("aa", ""));
ASSERT_OK(Put("xb", ""));
ASSERT_OK(Put("xx1", ""));
ASSERT_OK(Put("xz1", ""));
ASSERT_OK(Put("zz", ""));
ASSERT_OK(Flush());
// After reopening DB with prefix size 2 => 1, prefix extractor
// won't take effective unless it won't change results based
// on upper bound and seek key.
options.prefix_extractor.reset(NewFixedPrefixTransform(1));
Reopen(options);
{
std::unique_ptr<Iterator> iterator(db_->NewIterator(ReadOptions()));
ASSERT_OK(iterator->status());
iterator->Seek("xa");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("xb", iterator->key().ToString());
// It's a bug that the counter BLOOM_FILTER_PREFIX_CHECKED is not
// correct in this case. So don't check counters in this case.
if (expect_filter_check) {
ASSERT_EQ(0, TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED));
}
iterator->Seek("xz");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("xz1", iterator->key().ToString());
if (expect_filter_check) {
ASSERT_EQ(0, TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED));
}
}
std::string ub_str = "xg9";
Slice ub(ub_str);
ReadOptions ro;
ro.iterate_upper_bound = &ub;
{
std::unique_ptr<Iterator> iterator(db_->NewIterator(ro));
ASSERT_OK(iterator->status());
// SeekForPrev() never uses prefix bloom if it is changed.
iterator->SeekForPrev("xg0");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("xb", iterator->key().ToString());
if (expect_filter_check) {
ASSERT_EQ(0, TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED));
}
}
ub_str = "xx9";
ub = Slice(ub_str);
{
std::unique_ptr<Iterator> iterator(db_->NewIterator(ro));
ASSERT_OK(iterator->status());
iterator->Seek("x");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("xb", iterator->key().ToString());
if (expect_filter_check) {
ASSERT_EQ(0, TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED));
}
iterator->Seek("xx0");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("xx1", iterator->key().ToString());
if (expect_filter_check) {
ASSERT_EQ(1, TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED));
}
}
CompactRangeOptions compact_range_opts;
compact_range_opts.bottommost_level_compaction =
BottommostLevelCompaction::kForce;
ASSERT_OK(db_->CompactRange(compact_range_opts, nullptr, nullptr));
ASSERT_OK(db_->CompactRange(compact_range_opts, nullptr, nullptr));
// Re-execute similar queries after a full compaction
{
std::unique_ptr<Iterator> iterator(db_->NewIterator(ReadOptions()));
iterator->Seek("x");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("xb", iterator->key().ToString());
if (expect_filter_check) {
ASSERT_EQ(2, TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED));
}
iterator->Seek("xg");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("xx1", iterator->key().ToString());
if (expect_filter_check) {
ASSERT_EQ(3, TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED));
}
iterator->Seek("xz");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("xz1", iterator->key().ToString());
if (expect_filter_check) {
ASSERT_EQ(4, TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED));
}
ASSERT_OK(iterator->status());
}
{
std::unique_ptr<Iterator> iterator(db_->NewIterator(ro));
iterator->SeekForPrev("xx0");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("xb", iterator->key().ToString());
if (expect_filter_check) {
ASSERT_EQ(5, TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED));
}
iterator->Seek("xx0");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("xx1", iterator->key().ToString());
if (expect_filter_check) {
ASSERT_EQ(6, TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED));
}
ASSERT_OK(iterator->status());
}
ub_str = "xg9";
ub = Slice(ub_str);
{
std::unique_ptr<Iterator> iterator(db_->NewIterator(ro));
iterator->SeekForPrev("xg0");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("xb", iterator->key().ToString());
if (expect_filter_check) {
ASSERT_EQ(7, TestGetTickerCount(options, BLOOM_FILTER_PREFIX_CHECKED));
}
ASSERT_OK(iterator->status());
}
}
}
TEST_F(DBTest2, BlockBasedTablePrefixGetIndexNotFound) {
// create a DB with block prefix index
BlockBasedTableOptions table_options;
Options options = CurrentOptions();
table_options.block_size = 300;
table_options.index_type = BlockBasedTableOptions::kHashSearch;
table_options.index_shortening =
BlockBasedTableOptions::IndexShorteningMode::kNoShortening;
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.prefix_extractor.reset(NewFixedPrefixTransform(1));
options.level0_file_num_compaction_trigger = 8;
Reopen(options);
ASSERT_OK(Put("b1", "ok"));
ASSERT_OK(Flush());
// Flushing several files so that the chance that hash bucket
// is empty fo "b" in at least one of the files is high.
ASSERT_OK(Put("a1", ""));
ASSERT_OK(Put("c1", ""));
ASSERT_OK(Flush());
ASSERT_OK(Put("a2", ""));
ASSERT_OK(Put("c2", ""));
ASSERT_OK(Flush());
ASSERT_OK(Put("a3", ""));
ASSERT_OK(Put("c3", ""));
ASSERT_OK(Flush());
ASSERT_OK(Put("a4", ""));
ASSERT_OK(Put("c4", ""));
ASSERT_OK(Flush());
ASSERT_OK(Put("a5", ""));
ASSERT_OK(Put("c5", ""));
ASSERT_OK(Flush());
ASSERT_EQ("ok", Get("b1"));
}
#ifndef ROCKSDB_LITE
TEST_F(DBTest2, AutoPrefixMode1) {
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
do {
// create a DB with block prefix index
Options options = CurrentOptions();
BlockBasedTableOptions table_options =
*options.table_factory->GetOptions<BlockBasedTableOptions>();
table_options.filter_policy.reset(NewBloomFilterPolicy(10, false));
options.table_factory.reset(NewBlockBasedTableFactory(table_options));
options.prefix_extractor.reset(NewFixedPrefixTransform(1));
options.statistics = CreateDBStatistics();
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
Reopen(options);
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
Random rnd(301);
std::string large_value = rnd.RandomString(500);
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
ASSERT_OK(Put("a1", large_value));
ASSERT_OK(Put("x1", large_value));
ASSERT_OK(Put("y1", large_value));
ASSERT_OK(Flush());
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
ReadOptions ro;
ro.total_order_seek = false;
ro.auto_prefix_mode = true;
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
const auto stat = BLOOM_FILTER_PREFIX_CHECKED;
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
{
std::unique_ptr<Iterator> iterator(db_->NewIterator(ro));
iterator->Seek("b1");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("x1", iterator->key().ToString());
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
EXPECT_EQ(0, TestGetAndResetTickerCount(options, stat));
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
ASSERT_OK(iterator->status());
}
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
Slice ub;
ro.iterate_upper_bound = &ub;
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
ub = "b9";
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
{
std::unique_ptr<Iterator> iterator(db_->NewIterator(ro));
iterator->Seek("b1");
ASSERT_FALSE(iterator->Valid());
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
EXPECT_EQ(1, TestGetAndResetTickerCount(options, stat));
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
ASSERT_OK(iterator->status());
}
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
ub = "z";
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
{
std::unique_ptr<Iterator> iterator(db_->NewIterator(ro));
iterator->Seek("b1");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("x1", iterator->key().ToString());
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
EXPECT_EQ(0, TestGetAndResetTickerCount(options, stat));
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
ASSERT_OK(iterator->status());
}
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
ub = "c";
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
{
std::unique_ptr<Iterator> iterator(db_->NewIterator(ro));
iterator->Seek("b1");
ASSERT_FALSE(iterator->Valid());
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
EXPECT_EQ(1, TestGetAndResetTickerCount(options, stat));
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
ASSERT_OK(iterator->status());
}
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
ub = "c1";
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
{
std::unique_ptr<Iterator> iterator(db_->NewIterator(ro));
iterator->Seek("b1");
ASSERT_FALSE(iterator->Valid());
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
EXPECT_EQ(0, TestGetAndResetTickerCount(options, stat));
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
ASSERT_OK(iterator->status());
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
}
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
// The same queries without recreating iterator
{
std::unique_ptr<Iterator> iterator(db_->NewIterator(ro));
ub = "b9";
iterator->Seek("b1");
ASSERT_FALSE(iterator->Valid());
EXPECT_EQ(1, TestGetAndResetTickerCount(options, stat));
ASSERT_OK(iterator->status());
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
ub = "z";
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
iterator->Seek("b1");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("x1", iterator->key().ToString());
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
EXPECT_EQ(0, TestGetAndResetTickerCount(options, stat));
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
ub = "c";
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
iterator->Seek("b1");
ASSERT_FALSE(iterator->Valid());
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
EXPECT_EQ(1, TestGetAndResetTickerCount(options, stat));
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
ub = "b9";
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
iterator->SeekForPrev("b1");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("a1", iterator->key().ToString());
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
EXPECT_EQ(0, TestGetAndResetTickerCount(options, stat));
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
ub = "zz";
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
iterator->SeekToLast();
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("y1", iterator->key().ToString());
iterator->SeekToFirst();
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("a1", iterator->key().ToString());
}
Document design/specification bugs with auto_prefix_mode (#10144) Summary: auto_prefix_mode is designed to use prefix filtering in a particular "safe" set of cases where the upper bound and the seek key have different prefixes: where the upper bound is the "same length immediate successor". These conditions are not sufficient to guarantee the same iteration results as total_order_seek if the DB contains "short" keys, less than the "full" (maximum) prefix length. We are not simply disabling the optimization in these successor cases because it is likely that users are essentially getting what they want out of existing usage. Especially if users are constructing successor bounds with the intention of doing a prefix-bounded seek, the existing behavior is more expected than the total_order_seek behavior. Consequently, for now we reconcile the bad specification of behavior by documenting the existing mismatch with total_order_seek. A closely related issue affects hypothetical comparators like ReverseBytewiseComparator: if they "correctly" implement IsSameLengthImmediateSuccessor, auto_prefix_mode could omit more entries (other than "short" keys noted above). Luckily, the built-in ReverseBytewiseComparator has an "incorrect" implementation of IsSameLengthImmediateSuccessor that effectively prevents prefix optimization and, thus, the bug. This is now documented as a new constraint on IsSameLengthImmediateSuccessor, and the implementation tweaked to be simply "safe" rather than "incorrect". This change also includes unit test updates to demonstrate the above issues. (Test was cleaned up for readability and simplicity.) Intended follow-up: * Tweak documented axioms for prefix_extractor (more details then) * Consider some sort of fix for this case. I don't know what that would look like without breaking the performance of existing code. Perhaps if all keys in an SST file have prefixes that are "full length," we can track that fact and use it to allow optimization with the "same length immediate successor", but that would only apply to new files. * Consider a better system of specifying prefix bounds Pull Request resolved: https://github.com/facebook/rocksdb/pull/10144 Test Plan: test updates included Reviewed By: siying Differential Revision: D37052710 Pulled By: pdillinger fbshipit-source-id: 5f63b7d65f3f214e4b143e0f9aa1749527c587db
2 years ago
// Similar, now with reverse comparator
// Technically, we are violating axiom 2 of prefix_extractors, but
// it should be revised because of major use-cases using
// ReverseBytewiseComparator with capped/fixed prefix Seek. (FIXME)
options.comparator = ReverseBytewiseComparator();
options.prefix_extractor.reset(NewFixedPrefixTransform(1));
DestroyAndReopen(options);
ASSERT_OK(Put("a1", large_value));
ASSERT_OK(Put("x1", large_value));
ASSERT_OK(Put("y1", large_value));
ASSERT_OK(Flush());
{
std::unique_ptr<Iterator> iterator(db_->NewIterator(ro));
ub = "b1";
iterator->Seek("b9");
ASSERT_FALSE(iterator->Valid());
EXPECT_EQ(1, TestGetAndResetTickerCount(options, stat));
ASSERT_OK(iterator->status());
ub = "b1";
iterator->Seek("z");
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("y1", iterator->key().ToString());
EXPECT_EQ(0, TestGetAndResetTickerCount(options, stat));
ub = "b1";
iterator->Seek("c");
ASSERT_FALSE(iterator->Valid());
EXPECT_EQ(0, TestGetAndResetTickerCount(options, stat));
ub = "b";
iterator->Seek("c9");
ASSERT_FALSE(iterator->Valid());
// Fails if ReverseBytewiseComparator::IsSameLengthImmediateSuccessor
// is "correctly" implemented.
EXPECT_EQ(0, TestGetAndResetTickerCount(options, stat));
ub = "a";
iterator->Seek("b9");
// Fails if ReverseBytewiseComparator::IsSameLengthImmediateSuccessor
// is "correctly" implemented.
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("a1", iterator->key().ToString());
EXPECT_EQ(0, TestGetAndResetTickerCount(options, stat));
ub = "b";
iterator->Seek("a");
ASSERT_FALSE(iterator->Valid());
// Fails if ReverseBytewiseComparator::IsSameLengthImmediateSuccessor
// matches BytewiseComparator::IsSameLengthImmediateSuccessor. Upper
// comparing before seek key prevents a real bug from surfacing.
EXPECT_EQ(0, TestGetAndResetTickerCount(options, stat));
ub = "b1";
iterator->SeekForPrev("b9");
ASSERT_TRUE(iterator->Valid());
// Fails if ReverseBytewiseComparator::IsSameLengthImmediateSuccessor
// is "correctly" implemented.
ASSERT_EQ("x1", iterator->key().ToString());
EXPECT_EQ(0, TestGetAndResetTickerCount(options, stat));
ub = "a";
iterator->SeekToLast();
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("a1", iterator->key().ToString());
iterator->SeekToFirst();
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("y1", iterator->key().ToString());
}
// Now something a bit different, related to "short" keys that
// auto_prefix_mode can omit. See "BUG" section of auto_prefix_mode.
options.comparator = BytewiseComparator();
for (const auto config : {"fixed:2", "capped:2"}) {
ASSERT_OK(SliceTransform::CreateFromString(ConfigOptions(), config,
&options.prefix_extractor));
// FIXME: kHashSearch, etc. requires all keys be InDomain
if (StartsWith(config, "fixed") &&
(table_options.index_type == BlockBasedTableOptions::kHashSearch ||
StartsWith(options.memtable_factory->Name(), "Hash"))) {
continue;
}
DestroyAndReopen(options);
const char* a_end_stuff = "a\xffXYZ";
const char* b_begin_stuff = "b\x00XYZ";
ASSERT_OK(Put("a", large_value));
ASSERT_OK(Put("b", large_value));
ASSERT_OK(Put(Slice(b_begin_stuff, 3), large_value));
ASSERT_OK(Put("c", large_value));
ASSERT_OK(Flush());
// control showing valid optimization with auto_prefix mode
ub = Slice(a_end_stuff, 4);
ro.iterate_upper_bound = &ub;
std::unique_ptr<Iterator> iterator(db_->NewIterator(ro));
iterator->Seek(Slice(a_end_stuff, 2));
ASSERT_FALSE(iterator->Valid());
EXPECT_EQ(1, TestGetAndResetTickerCount(options, stat));
ASSERT_OK(iterator->status());
// test, cannot be validly optimized with auto_prefix_mode
ub = Slice(b_begin_stuff, 2);
ro.iterate_upper_bound = &ub;
iterator->Seek(Slice(a_end_stuff, 2));
// !!! BUG !!! See "BUG" section of auto_prefix_mode.
ASSERT_FALSE(iterator->Valid());
EXPECT_EQ(1, TestGetAndResetTickerCount(options, stat));
ASSERT_OK(iterator->status());
// To prove that is the wrong result, now use total order seek
ReadOptions tos_ro = ro;
tos_ro.total_order_seek = true;
tos_ro.auto_prefix_mode = false;
iterator.reset(db_->NewIterator(tos_ro));
iterator->Seek(Slice(a_end_stuff, 2));
ASSERT_TRUE(iterator->Valid());
ASSERT_EQ("b", iterator->key().ToString());
EXPECT_EQ(0, TestGetAndResetTickerCount(options, stat));
ASSERT_OK(iterator->status());
}
Fix auto_prefix_mode performance with partitioned filters (#10012) Summary: Essentially refactored the RangeMayExist implementation in FullFilterBlockReader to FilterBlockReaderCommon so that it applies to partitioned filters as well. (The function is not called for the block-based filter case.) RangeMayExist is essentially a series of checks around a possible PrefixMayExist, and I'm confident those checks should be the same for partitioned as for full filters. (I think it's likely that bugs remain in those checks, but this change is overall a simplifying one.) Added auto_prefix_mode support to db_bench Other small fixes as well Fixes https://github.com/facebook/rocksdb/issues/10003 Pull Request resolved: https://github.com/facebook/rocksdb/pull/10012 Test Plan: Expanded unit test that uses statistics to check for filter optimization, fails without the production code changes here Performance: populate two DBs with ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=30000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters ``` Observe no measurable change in non-partitioned performance ``` TEST_TMPDIR=/dev/shm/rocksdb_nonpartitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 15 runs] : 11798 (± 331) ops/sec After: seekrandom [AVG 15 runs] : 11724 (± 315) ops/sec Observe big improvement with partitioned (also supported by bloom use statistics) ``` TEST_TMPDIR=/dev/shm/rocksdb_partitioned ./db_bench -benchmarks=seekrandom[-X1000] -num=10000000 -readonly -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=8 -partition_index_and_filters -auto_prefix_mode -cache_index_and_filter_blocks=1 -cache_size=1000000000 -duration 20 ``` Before: seekrandom [AVG 12 runs] : 2942 (± 57) ops/sec After: seekrandom [AVG 12 runs] : 7489 (± 184) ops/sec Reviewed By: siying Differential Revision: D36469796 Pulled By: pdillinger fbshipit-source-id: bcf1e2a68d347b32adb2b27384f945434e7a266d
3 years ago
} while (ChangeOptions(kSkipPlainTable));
}
Handle rename() failure in non-local FS (#8192) Summary: In a distributed environment, a file `rename()` operation can succeed on server (remote) side, but the client can somehow return non-ok status to RocksDB. Possible reasons include network partition, connection issue, etc. This happens in `rocksdb::SetCurrentFile()`, which can be called in `LogAndApply() -> ProcessManifestWrites()` if RocksDB tries to switch to a new MANIFEST. We currently always delete the new MANIFEST if an error occurs. This is problematic in distributed world. If the server-side successfully updates the CURRENT file via renaming, then a subsequent `DB::Open()` will try to look for the new MANIFEST and fail. As a fix, we can track the execution result of IO operations on the new MANIFEST. - If IO operations on the new MANIFEST fail, then we know the CURRENT must point to the original MANIFEST. Therefore, it is safe to remove the new MANIFEST. - If IO operations on the new MANIFEST all succeed, but somehow we end up in the clean up code block, then we do not know whether CURRENT points to the new or old MANIFEST. (For local POSIX-compliant FS, it should still point to old MANIFEST, but it does not matter if we keep the new MANIFEST.) Therefore, we keep the new MANIFEST. - Any future `LogAndApply()` will switch to a new MANIFEST and update CURRENT. - If process reopens the db immediately after the failure, then the CURRENT file can point to either the new MANIFEST or the old one, both of which exist. Therefore, recovery can succeed and ignore the other. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8192 Test Plan: make check Reviewed By: zhichao-cao Differential Revision: D27804648 Pulled By: riversand963 fbshipit-source-id: 9c16f2a5ce41bc6aadf085e48449b19ede8423e4
4 years ago
class RenameCurrentTest : public DBTestBase,
public testing::WithParamInterface<std::string> {
public:
RenameCurrentTest()
: DBTestBase("rename_current_test", /*env_do_fsync=*/true),
sync_point_(GetParam()) {}
~RenameCurrentTest() override {}
void SetUp() override {
env_->no_file_overwrite_.store(true, std::memory_order_release);
}
void TearDown() override {
env_->no_file_overwrite_.store(false, std::memory_order_release);
}
void SetupSyncPoints() {
SyncPoint::GetInstance()->DisableProcessing();
SyncPoint::GetInstance()->SetCallBack(sync_point_, [&](void* arg) {
Status* s = reinterpret_cast<Status*>(arg);
assert(s);
*s = Status::IOError("Injected IO error.");
});
}
const std::string sync_point_;
};
INSTANTIATE_TEST_CASE_P(DistributedFS, RenameCurrentTest,
::testing::Values("SetCurrentFile:BeforeRename",
"SetCurrentFile:AfterRename"));
TEST_P(RenameCurrentTest, Open) {
Destroy(last_options_);
Options options = GetDefaultOptions();
options.create_if_missing = true;
SetupSyncPoints();
SyncPoint::GetInstance()->EnableProcessing();
Status s = TryReopen(options);
ASSERT_NOK(s);
SyncPoint::GetInstance()->DisableProcessing();
Reopen(options);
}
TEST_P(RenameCurrentTest, Flush) {
Destroy(last_options_);
Options options = GetDefaultOptions();
options.max_manifest_file_size = 1;
options.create_if_missing = true;
Reopen(options);
ASSERT_OK(Put("key", "value"));
SetupSyncPoints();
SyncPoint::GetInstance()->EnableProcessing();
ASSERT_NOK(Flush());
ASSERT_NOK(Put("foo", "value"));
SyncPoint::GetInstance()->DisableProcessing();
Reopen(options);
ASSERT_EQ("value", Get("key"));
ASSERT_EQ("NOT_FOUND", Get("foo"));
}
TEST_P(RenameCurrentTest, Compaction) {
Destroy(last_options_);
Options options = GetDefaultOptions();
options.max_manifest_file_size = 1;
options.create_if_missing = true;
Reopen(options);
ASSERT_OK(Put("a", "a_value"));
ASSERT_OK(Put("c", "c_value"));
ASSERT_OK(Flush());
ASSERT_OK(Put("b", "b_value"));
ASSERT_OK(Put("d", "d_value"));
ASSERT_OK(Flush());
SetupSyncPoints();
SyncPoint::GetInstance()->EnableProcessing();
ASSERT_NOK(db_->CompactRange(CompactRangeOptions(), /*begin=*/nullptr,
/*end=*/nullptr));
ASSERT_NOK(Put("foo", "value"));
SyncPoint::GetInstance()->DisableProcessing();
Reopen(options);
ASSERT_EQ("NOT_FOUND", Get("foo"));
ASSERT_EQ("d_value", Get("d"));
}
TEST_F(DBTest2, LastLevelTemperature) {
class TestListener : public EventListener {
public:
void OnFileReadFinish(const FileOperationInfo& info) override {
UpdateFileTemperature(info);
}
void OnFileWriteFinish(const FileOperationInfo& info) override {
UpdateFileTemperature(info);
}
void OnFileFlushFinish(const FileOperationInfo& info) override {
UpdateFileTemperature(info);
}
void OnFileSyncFinish(const FileOperationInfo& info) override {
UpdateFileTemperature(info);
}
void OnFileCloseFinish(const FileOperationInfo& info) override {
UpdateFileTemperature(info);
}
bool ShouldBeNotifiedOnFileIO() override { return true; }
std::unordered_map<uint64_t, Temperature> file_temperatures;
private:
void UpdateFileTemperature(const FileOperationInfo& info) {
auto filename = GetFileName(info.path);
uint64_t number;
FileType type;
ASSERT_TRUE(ParseFileName(filename, &number, &type));
if (type == kTableFile) {
MutexLock l(&mutex_);
auto ret = file_temperatures.insert({number, info.temperature});
if (!ret.second) {
// the same file temperature should always be the same for all events
ASSERT_TRUE(ret.first->second == info.temperature);
}
}
}
std::string GetFileName(const std::string& fname) {
auto filename = fname.substr(fname.find_last_of(kFilePathSeparator) + 1);
// workaround only for Windows that the file path could contain both
// Windows FilePathSeparator and '/'
filename = filename.substr(filename.find_last_of('/') + 1);
return filename;
}
port::Mutex mutex_;
};
const int kNumLevels = 7;
const int kLastLevel = kNumLevels - 1;
auto* listener = new TestListener();
Options options = CurrentOptions();
options.bottommost_temperature = Temperature::kWarm;
options.level0_file_num_compaction_trigger = 2;
options.level_compaction_dynamic_level_bytes = true;
options.num_levels = kNumLevels;
options.statistics = CreateDBStatistics();
options.listeners.emplace_back(listener);
Reopen(options);
auto size = GetSstSizeHelper(Temperature::kUnknown);
ASSERT_EQ(size, 0);
size = GetSstSizeHelper(Temperature::kWarm);
ASSERT_EQ(size, 0);
size = GetSstSizeHelper(Temperature::kHot);
ASSERT_EQ(size, 0);
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Put("bar", "bar"));
ASSERT_OK(Flush());
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Put("bar", "bar"));
ASSERT_OK(Flush());
ASSERT_OK(dbfull()->TEST_WaitForCompact());
get_iostats_context()->Reset();
IOStatsContext* iostats = get_iostats_context();
ColumnFamilyMetaData metadata;
db_->GetColumnFamilyMetaData(&metadata);
ASSERT_EQ(1, metadata.file_count);
SstFileMetaData meta = metadata.levels[kLastLevel].files[0];
ASSERT_EQ(Temperature::kWarm, meta.temperature);
uint64_t number;
FileType type;
ASSERT_TRUE(ParseFileName(meta.name, &number, &type));
ASSERT_EQ(listener->file_temperatures.at(number), meta.temperature);
size = GetSstSizeHelper(Temperature::kUnknown);
ASSERT_EQ(size, 0);
size = GetSstSizeHelper(Temperature::kWarm);
ASSERT_GT(size, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.hot_file_read_count, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.warm_file_read_count, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.hot_file_read_count, 0);
ASSERT_EQ(options.statistics->getTickerCount(HOT_FILE_READ_BYTES), 0);
ASSERT_GT(options.statistics->getTickerCount(WARM_FILE_READ_BYTES), 0);
ASSERT_EQ(options.statistics->getTickerCount(COLD_FILE_READ_BYTES), 0);
ASSERT_EQ("bar", Get("foo"));
ASSERT_EQ(iostats->file_io_stats_by_temperature.hot_file_read_count, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.warm_file_read_count, 1);
ASSERT_EQ(iostats->file_io_stats_by_temperature.hot_file_read_count, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.hot_file_bytes_read, 0);
ASSERT_GT(iostats->file_io_stats_by_temperature.warm_file_bytes_read, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.cold_file_bytes_read, 0);
ASSERT_EQ(options.statistics->getTickerCount(HOT_FILE_READ_BYTES), 0);
ASSERT_GT(options.statistics->getTickerCount(WARM_FILE_READ_BYTES), 0);
ASSERT_EQ(options.statistics->getTickerCount(COLD_FILE_READ_BYTES), 0);
ASSERT_EQ(options.statistics->getTickerCount(HOT_FILE_READ_COUNT), 0);
ASSERT_GT(options.statistics->getTickerCount(WARM_FILE_READ_COUNT), 0);
ASSERT_EQ(options.statistics->getTickerCount(COLD_FILE_READ_COUNT), 0);
// non-bottommost file still has unknown temperature
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Put("bar", "bar"));
ASSERT_OK(Flush());
ASSERT_EQ("bar", Get("bar"));
ASSERT_EQ(iostats->file_io_stats_by_temperature.hot_file_read_count, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.warm_file_read_count, 1);
ASSERT_EQ(iostats->file_io_stats_by_temperature.hot_file_read_count, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.hot_file_bytes_read, 0);
ASSERT_GT(iostats->file_io_stats_by_temperature.warm_file_bytes_read, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.cold_file_bytes_read, 0);
ASSERT_EQ(options.statistics->getTickerCount(HOT_FILE_READ_BYTES), 0);
ASSERT_GT(options.statistics->getTickerCount(WARM_FILE_READ_BYTES), 0);
ASSERT_EQ(options.statistics->getTickerCount(COLD_FILE_READ_BYTES), 0);
ASSERT_EQ(options.statistics->getTickerCount(HOT_FILE_READ_COUNT), 0);
ASSERT_GT(options.statistics->getTickerCount(WARM_FILE_READ_COUNT), 0);
ASSERT_EQ(options.statistics->getTickerCount(COLD_FILE_READ_COUNT), 0);
db_->GetColumnFamilyMetaData(&metadata);
ASSERT_EQ(2, metadata.file_count);
meta = metadata.levels[0].files[0];
ASSERT_EQ(Temperature::kUnknown, meta.temperature);
ASSERT_TRUE(ParseFileName(meta.name, &number, &type));
ASSERT_EQ(listener->file_temperatures.at(number), meta.temperature);
meta = metadata.levels[kLastLevel].files[0];
ASSERT_EQ(Temperature::kWarm, meta.temperature);
ASSERT_TRUE(ParseFileName(meta.name, &number, &type));
ASSERT_EQ(listener->file_temperatures.at(number), meta.temperature);
size = GetSstSizeHelper(Temperature::kUnknown);
ASSERT_GT(size, 0);
size = GetSstSizeHelper(Temperature::kWarm);
ASSERT_GT(size, 0);
// reopen and check the information is persisted
Reopen(options);
db_->GetColumnFamilyMetaData(&metadata);
ASSERT_EQ(2, metadata.file_count);
meta = metadata.levels[0].files[0];
ASSERT_EQ(Temperature::kUnknown, meta.temperature);
ASSERT_TRUE(ParseFileName(meta.name, &number, &type));
ASSERT_EQ(listener->file_temperatures.at(number), meta.temperature);
meta = metadata.levels[kLastLevel].files[0];
ASSERT_EQ(Temperature::kWarm, meta.temperature);
ASSERT_TRUE(ParseFileName(meta.name, &number, &type));
ASSERT_EQ(listener->file_temperatures.at(number), meta.temperature);
size = GetSstSizeHelper(Temperature::kUnknown);
ASSERT_GT(size, 0);
size = GetSstSizeHelper(Temperature::kWarm);
ASSERT_GT(size, 0);
// check other non-exist temperatures
size = GetSstSizeHelper(Temperature::kHot);
ASSERT_EQ(size, 0);
size = GetSstSizeHelper(Temperature::kCold);
ASSERT_EQ(size, 0);
std::string prop;
ASSERT_TRUE(dbfull()->GetProperty(
DB::Properties::kLiveSstFilesSizeAtTemperature + std::to_string(22),
&prop));
ASSERT_EQ(std::atoi(prop.c_str()), 0);
Reopen(options);
db_->GetColumnFamilyMetaData(&metadata);
ASSERT_EQ(2, metadata.file_count);
meta = metadata.levels[0].files[0];
ASSERT_EQ(Temperature::kUnknown, meta.temperature);
ASSERT_TRUE(ParseFileName(meta.name, &number, &type));
ASSERT_EQ(listener->file_temperatures.at(number), meta.temperature);
meta = metadata.levels[kLastLevel].files[0];
ASSERT_EQ(Temperature::kWarm, meta.temperature);
ASSERT_TRUE(ParseFileName(meta.name, &number, &type));
ASSERT_EQ(listener->file_temperatures.at(number), meta.temperature);
}
TEST_F(DBTest2, LastLevelTemperatureUniversal) {
const int kTriggerNum = 3;
const int kNumLevels = 5;
const int kBottommostLevel = kNumLevels - 1;
Options options = CurrentOptions();
options.compaction_style = kCompactionStyleUniversal;
options.level0_file_num_compaction_trigger = kTriggerNum;
options.num_levels = kNumLevels;
options.statistics = CreateDBStatistics();
DestroyAndReopen(options);
auto size = GetSstSizeHelper(Temperature::kUnknown);
ASSERT_EQ(size, 0);
size = GetSstSizeHelper(Temperature::kWarm);
ASSERT_EQ(size, 0);
size = GetSstSizeHelper(Temperature::kHot);
ASSERT_EQ(size, 0);
get_iostats_context()->Reset();
IOStatsContext* iostats = get_iostats_context();
for (int i = 0; i < kTriggerNum; i++) {
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Put("bar", "bar"));
ASSERT_OK(Flush());
}
ASSERT_OK(dbfull()->TEST_WaitForCompact());
ColumnFamilyMetaData metadata;
db_->GetColumnFamilyMetaData(&metadata);
ASSERT_EQ(1, metadata.file_count);
ASSERT_EQ(Temperature::kUnknown,
metadata.levels[kBottommostLevel].files[0].temperature);
size = GetSstSizeHelper(Temperature::kUnknown);
ASSERT_GT(size, 0);
size = GetSstSizeHelper(Temperature::kWarm);
ASSERT_EQ(size, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.hot_file_read_count, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.warm_file_read_count, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.hot_file_read_count, 0);
ASSERT_EQ(options.statistics->getTickerCount(HOT_FILE_READ_BYTES), 0);
ASSERT_EQ(options.statistics->getTickerCount(WARM_FILE_READ_BYTES), 0);
ASSERT_EQ(options.statistics->getTickerCount(COLD_FILE_READ_BYTES), 0);
ASSERT_EQ(options.statistics->getTickerCount(HOT_FILE_READ_COUNT), 0);
ASSERT_EQ(options.statistics->getTickerCount(WARM_FILE_READ_COUNT), 0);
ASSERT_EQ(options.statistics->getTickerCount(COLD_FILE_READ_COUNT), 0);
ASSERT_EQ("bar", Get("foo"));
ASSERT_EQ(iostats->file_io_stats_by_temperature.hot_file_read_count, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.warm_file_read_count, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.hot_file_read_count, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.hot_file_bytes_read, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.warm_file_bytes_read, 0);
ASSERT_EQ(iostats->file_io_stats_by_temperature.cold_file_bytes_read, 0);
ASSERT_EQ(options.statistics->getTickerCount(HOT_FILE_READ_BYTES), 0);
ASSERT_EQ(options.statistics->getTickerCount(WARM_FILE_READ_BYTES), 0);
ASSERT_EQ(options.statistics->getTickerCount(COLD_FILE_READ_BYTES), 0);
ASSERT_EQ(options.statistics->getTickerCount(HOT_FILE_READ_COUNT), 0);
ASSERT_EQ(options.statistics->getTickerCount(WARM_FILE_READ_COUNT), 0);
ASSERT_EQ(options.statistics->getTickerCount(COLD_FILE_READ_COUNT), 0);
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Put("bar", "bar"));
ASSERT_OK(Flush());
ASSERT_OK(dbfull()->TEST_WaitForCompact());
db_->GetColumnFamilyMetaData(&metadata);
ASSERT_EQ(2, metadata.file_count);
ASSERT_EQ(Temperature::kUnknown, metadata.levels[0].files[0].temperature);
size = GetSstSizeHelper(Temperature::kUnknown);
ASSERT_GT(size, 0);
size = GetSstSizeHelper(Temperature::kWarm);
ASSERT_EQ(size, 0);
// Update bottommost temperature
options.bottommost_temperature = Temperature::kWarm;
Reopen(options);
db_->GetColumnFamilyMetaData(&metadata);
// Should not impact existing ones
ASSERT_EQ(Temperature::kUnknown,
metadata.levels[kBottommostLevel].files[0].temperature);
size = GetSstSizeHelper(Temperature::kUnknown);
ASSERT_GT(size, 0);
size = GetSstSizeHelper(Temperature::kWarm);
ASSERT_EQ(size, 0);
// new generated file should have the new settings
ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
db_->GetColumnFamilyMetaData(&metadata);
ASSERT_EQ(1, metadata.file_count);
ASSERT_EQ(Temperature::kWarm,
metadata.levels[kBottommostLevel].files[0].temperature);
size = GetSstSizeHelper(Temperature::kUnknown);
ASSERT_EQ(size, 0);
size = GetSstSizeHelper(Temperature::kWarm);
ASSERT_GT(size, 0);
ASSERT_EQ(options.statistics->getTickerCount(HOT_FILE_READ_BYTES), 0);
ASSERT_GT(options.statistics->getTickerCount(WARM_FILE_READ_BYTES), 0);
ASSERT_EQ(options.statistics->getTickerCount(COLD_FILE_READ_BYTES), 0);
ASSERT_EQ(options.statistics->getTickerCount(HOT_FILE_READ_COUNT), 0);
ASSERT_GT(options.statistics->getTickerCount(WARM_FILE_READ_COUNT), 0);
ASSERT_EQ(options.statistics->getTickerCount(COLD_FILE_READ_COUNT), 0);
// non-bottommost file still has unknown temperature
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Put("bar", "bar"));
ASSERT_OK(Flush());
ASSERT_OK(dbfull()->TEST_WaitForCompact());
db_->GetColumnFamilyMetaData(&metadata);
ASSERT_EQ(2, metadata.file_count);
ASSERT_EQ(Temperature::kUnknown, metadata.levels[0].files[0].temperature);
size = GetSstSizeHelper(Temperature::kUnknown);
ASSERT_GT(size, 0);
size = GetSstSizeHelper(Temperature::kWarm);
ASSERT_GT(size, 0);
// check other non-exist temperatures
size = GetSstSizeHelper(Temperature::kHot);
ASSERT_EQ(size, 0);
size = GetSstSizeHelper(Temperature::kCold);
ASSERT_EQ(size, 0);
std::string prop;
ASSERT_TRUE(dbfull()->GetProperty(
DB::Properties::kLiveSstFilesSizeAtTemperature + std::to_string(22),
&prop));
ASSERT_EQ(std::atoi(prop.c_str()), 0);
// Update bottommost temperature dynamically with SetOptions
auto s = db_->SetOptions({{"last_level_temperature", "kCold"}});
ASSERT_OK(s);
ASSERT_EQ(db_->GetOptions().bottommost_temperature, Temperature::kCold);
db_->GetColumnFamilyMetaData(&metadata);
// Should not impact the existing files
ASSERT_EQ(Temperature::kWarm,
metadata.levels[kBottommostLevel].files[0].temperature);
size = GetSstSizeHelper(Temperature::kUnknown);
ASSERT_GT(size, 0);
size = GetSstSizeHelper(Temperature::kWarm);
ASSERT_GT(size, 0);
size = GetSstSizeHelper(Temperature::kCold);
ASSERT_EQ(size, 0);
// new generated files should have the new settings
ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
db_->GetColumnFamilyMetaData(&metadata);
ASSERT_EQ(1, metadata.file_count);
ASSERT_EQ(Temperature::kCold,
metadata.levels[kBottommostLevel].files[0].temperature);
size = GetSstSizeHelper(Temperature::kUnknown);
ASSERT_EQ(size, 0);
size = GetSstSizeHelper(Temperature::kWarm);
ASSERT_EQ(size, 0);
size = GetSstSizeHelper(Temperature::kCold);
ASSERT_GT(size, 0);
// kLastTemperature is an invalid temperature
options.bottommost_temperature = Temperature::kLastTemperature;
s = TryReopen(options);
ASSERT_TRUE(s.IsIOError());
}
TEST_F(DBTest2, LastLevelStatistics) {
Options options = CurrentOptions();
options.bottommost_temperature = Temperature::kWarm;
options.level0_file_num_compaction_trigger = 2;
options.level_compaction_dynamic_level_bytes = true;
options.statistics = CreateDBStatistics();
Reopen(options);
// generate 1 sst on level 0
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Put("bar", "bar"));
ASSERT_OK(Flush());
ASSERT_EQ("bar", Get("bar"));
ASSERT_GT(options.statistics->getTickerCount(NON_LAST_LEVEL_READ_BYTES), 0);
ASSERT_GT(options.statistics->getTickerCount(NON_LAST_LEVEL_READ_COUNT), 0);
ASSERT_EQ(options.statistics->getTickerCount(LAST_LEVEL_READ_BYTES), 0);
ASSERT_EQ(options.statistics->getTickerCount(LAST_LEVEL_READ_COUNT), 0);
// 2nd flush to trigger compaction
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Put("bar", "bar"));
ASSERT_OK(Flush());
ASSERT_OK(dbfull()->TEST_WaitForCompact());
ASSERT_EQ("bar", Get("bar"));
ASSERT_EQ(options.statistics->getTickerCount(LAST_LEVEL_READ_BYTES),
options.statistics->getTickerCount(WARM_FILE_READ_BYTES));
ASSERT_EQ(options.statistics->getTickerCount(LAST_LEVEL_READ_COUNT),
options.statistics->getTickerCount(WARM_FILE_READ_COUNT));
auto pre_bytes =
options.statistics->getTickerCount(NON_LAST_LEVEL_READ_BYTES);
auto pre_count =
options.statistics->getTickerCount(NON_LAST_LEVEL_READ_COUNT);
// 3rd flush to generate 1 sst on level 0
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Put("bar", "bar"));
ASSERT_OK(Flush());
ASSERT_EQ("bar", Get("bar"));
ASSERT_GT(options.statistics->getTickerCount(NON_LAST_LEVEL_READ_BYTES),
pre_bytes);
ASSERT_GT(options.statistics->getTickerCount(NON_LAST_LEVEL_READ_COUNT),
pre_count);
ASSERT_EQ(options.statistics->getTickerCount(LAST_LEVEL_READ_BYTES),
options.statistics->getTickerCount(WARM_FILE_READ_BYTES));
ASSERT_EQ(options.statistics->getTickerCount(LAST_LEVEL_READ_COUNT),
options.statistics->getTickerCount(WARM_FILE_READ_COUNT));
}
TEST_F(DBTest2, CheckpointFileTemperature) {
class NoLinkTestFS : public FileTemperatureTestFS {
using FileTemperatureTestFS::FileTemperatureTestFS;
Fix DBTest2.BackupFileTemperature memory leak (#9610) Summary: Valgrind was failing with the below error because we forgot to destroy the `BackupEngine` object: ``` ==421173== Command: ./db_test2 --gtest_filter=DBTest2.BackupFileTemperature ==421173== Note: Google Test filter = DBTest2.BackupFileTemperature [==========] Running 1 test from 1 test case. [----------] Global test environment set-up. [----------] 1 test from DBTest2 [ RUN ] DBTest2.BackupFileTemperature --421173-- WARNING: unhandled amd64-linux syscall: 425 --421173-- You may be able to write your own handler. --421173-- Read the file README_MISSING_SYSCALL_OR_IOCTL. --421173-- Nevertheless we consider this a bug. Please report --421173-- it at http://valgrind.org/support/bug_reports.html. [ OK ] DBTest2.BackupFileTemperature (3366 ms) [----------] 1 test from DBTest2 (3371 ms total) [----------] Global test environment tear-down [==========] 1 test from 1 test case ran. (3413 ms total) [ PASSED ] 1 test. ==421173== ==421173== HEAP SUMMARY: ==421173== in use at exit: 13,042 bytes in 195 blocks ==421173== total heap usage: 26,022 allocs, 25,827 frees, 27,555,265 bytes allocated ==421173== ==421173== 8 bytes in 1 blocks are possibly lost in loss record 6 of 167 ==421173== at 0x4838DBF: operator new(unsigned long) (vg_replace_malloc.c:344) ==421173== by 0x8D4606: allocate (new_allocator.h:114) ==421173== by 0x8D4606: allocate (alloc_traits.h:445) ==421173== by 0x8D4606: _M_allocate (stl_vector.h:343) ==421173== by 0x8D4606: reserve (vector.tcc:78) ==421173== by 0x8D4606: rocksdb::BackupEngineImpl::Initialize() (backupable_db.cc:1174) ==421173== by 0x8D5473: Initialize (backupable_db.cc:918) ==421173== by 0x8D5473: rocksdb::BackupEngine::Open(rocksdb::BackupEngineOptions const&, rocksdb::Env*, rocksdb::BackupEngine**) (backupable_db.cc:937) ==421173== by 0x50AC8F: Open (backup_engine.h:585) ==421173== by 0x50AC8F: rocksdb::DBTest2_BackupFileTemperature_Test::TestBody() (db_test2.cc:6996) ... ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/9610 Test Plan: ``` $ make -j24 ROCKSDBTESTS_SUBSET=db_test2 valgrind_check_some ``` Reviewed By: akankshamahajan15 Differential Revision: D34371210 Pulled By: ajkr fbshipit-source-id: 68154fcb0c51b28222efa23fa4ee02df8d925a18
3 years ago
IOStatus LinkFile(const std::string&, const std::string&, const IOOptions&,
IODebugContext*) override {
// return not supported to force checkpoint copy the file instead of just
// link
return IOStatus::NotSupported();
}
};
auto test_fs = std::make_shared<NoLinkTestFS>(env_->GetFileSystem());
std::unique_ptr<Env> env(new CompositeEnvWrapper(env_, test_fs));
Options options = CurrentOptions();
options.bottommost_temperature = Temperature::kWarm;
// set dynamic_level to true so the compaction would compact the data to the
// last level directly which will have the last_level_temperature
options.level_compaction_dynamic_level_bytes = true;
options.level0_file_num_compaction_trigger = 2;
options.env = env.get();
Reopen(options);
// generate a bottommost file and a non-bottommost file
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Put("bar", "bar"));
ASSERT_OK(Flush());
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Put("bar", "bar"));
ASSERT_OK(Flush());
ASSERT_OK(dbfull()->TEST_WaitForCompact());
ASSERT_OK(Put("foo", "bar"));
ASSERT_OK(Put("bar", "bar"));
ASSERT_OK(Flush());
auto size = GetSstSizeHelper(Temperature::kWarm);
ASSERT_GT(size, 0);
std::map<uint64_t, Temperature> temperatures;
std::vector<LiveFileStorageInfo> infos;
ASSERT_OK(
dbfull()->GetLiveFilesStorageInfo(LiveFilesStorageInfoOptions(), &infos));
for (auto info : infos) {
temperatures.emplace(info.file_number, info.temperature);
}
New backup meta schema, with file temperatures (#9660) Summary: The primary goal of this change is to add support for backing up and restoring (applying on restore) file temperature metadata, without committing to either the DB manifest or the FS reported "current" temperatures being exclusive "source of truth". To achieve this goal, we need to add temperature information to backup metadata, which requires updated backup meta schema. Fortunately I prepared for this in https://github.com/facebook/rocksdb/issues/8069, which began forward compatibility in version 6.19.0 for this kind of schema update. (Previously, backup meta schema was not extensible! Making this schema update public will allow some other "nice to have" features like taking backups with hard links, and avoiding crc32c checksum computation when another checksum is already available.) While schema version 2 is newly public, the default schema version is still 1. Until we change the default, users will need to set to 2 to enable features like temperature data backup+restore. New metadata like temperature information will be ignored with a warning in versions before this change and since 6.19.0. The metadata is considered ignorable because a functioning DB can be restored without it. Some detail: * Some renaming because "future schema" is now just public schema 2. * Initialize some atomics in TestFs (linter reported) * Add temperature hint support to SstFileDumper (used by BackupEngine) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9660 Test Plan: related unit test majorly updated for the new functionality, including some shared testing support for tracking temperatures in a FS. Some other tests and testing hooks into production code also updated for making the backup meta schema change public. Reviewed By: ajkr Differential Revision: D34686968 Pulled By: pdillinger fbshipit-source-id: 3ac1fa3e67ee97ca8a5103d79cc87d872c1d862a
3 years ago
test_fs->PopRequestedSstFileTemperatures();
Checkpoint* checkpoint;
ASSERT_OK(Checkpoint::Create(db_, &checkpoint));
ASSERT_OK(
checkpoint->CreateCheckpoint(dbname_ + kFilePathSeparator + "tempcp"));
// checking src file src_temperature hints: 2 sst files: 1 sst is kWarm,
// another is kUnknown
New backup meta schema, with file temperatures (#9660) Summary: The primary goal of this change is to add support for backing up and restoring (applying on restore) file temperature metadata, without committing to either the DB manifest or the FS reported "current" temperatures being exclusive "source of truth". To achieve this goal, we need to add temperature information to backup metadata, which requires updated backup meta schema. Fortunately I prepared for this in https://github.com/facebook/rocksdb/issues/8069, which began forward compatibility in version 6.19.0 for this kind of schema update. (Previously, backup meta schema was not extensible! Making this schema update public will allow some other "nice to have" features like taking backups with hard links, and avoiding crc32c checksum computation when another checksum is already available.) While schema version 2 is newly public, the default schema version is still 1. Until we change the default, users will need to set to 2 to enable features like temperature data backup+restore. New metadata like temperature information will be ignored with a warning in versions before this change and since 6.19.0. The metadata is considered ignorable because a functioning DB can be restored without it. Some detail: * Some renaming because "future schema" is now just public schema 2. * Initialize some atomics in TestFs (linter reported) * Add temperature hint support to SstFileDumper (used by BackupEngine) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9660 Test Plan: related unit test majorly updated for the new functionality, including some shared testing support for tracking temperatures in a FS. Some other tests and testing hooks into production code also updated for making the backup meta schema change public. Reviewed By: ajkr Differential Revision: D34686968 Pulled By: pdillinger fbshipit-source-id: 3ac1fa3e67ee97ca8a5103d79cc87d872c1d862a
3 years ago
std::vector<std::pair<uint64_t, Temperature>> requested_temps;
test_fs->PopRequestedSstFileTemperatures(&requested_temps);
// Two requests
ASSERT_EQ(requested_temps.size(), 2);
std::set<uint64_t> distinct_requests;
for (const auto& requested_temp : requested_temps) {
// Matching manifest temperatures
ASSERT_EQ(temperatures.at(requested_temp.first), requested_temp.second);
distinct_requests.insert(requested_temp.first);
}
// Each request to distinct file
ASSERT_EQ(distinct_requests.size(), requested_temps.size());
delete checkpoint;
Close();
}
Add manifest fix-up utility for file temperatures (#9683) Summary: The goal of this change is to allow changes to the "current" (in FileSystem) file temperatures to feed back into DB metadata, so that they can inform decisions and stats reporting. In part because of modular code factoring, it doesn't seem easy to do this automagically, where opening an SST file and observing current Temperature different from expected would trigger a change in metadata and DB manifest write (essentially giving the deep read path access to the write path). It is also difficult to do this while the DB is open because of the limitations of LogAndApply. This change allows updating file temperature metadata on a closed DB using an experimental utility function UpdateManifestForFilesState() or `ldb update_manifest --update_temperatures`. This should suffice for "migration" scenarios where outside tooling has placed or re-arranged DB files into a (different) tiered configuration without going through RocksDB itself (currently, only compaction can change temperature metadata). Some details: * Refactored and added unit test for `ldb unsafe_remove_sst_file` because of shared functionality * Pulled in autovector.h changes from https://github.com/facebook/rocksdb/issues/9546 to fix SuperVersionContext move constructor (related to an older draft of this change) Possible follow-up work: * Support updating manifest with file checksums, such as when a new checksum function is used and want existing DB metadata updated for it. * It's possible that for some repair scenarios, lighter weight than full repair, we might want to support UpdateManifestForFilesState() to modify critical file details like size or checksum using same algorithm. But let's make sure these are differentiated from modifying file details in ways that don't suspect corruption (or require extreme trust). Pull Request resolved: https://github.com/facebook/rocksdb/pull/9683 Test Plan: unit tests added Reviewed By: jay-zhuang Differential Revision: D34798828 Pulled By: pdillinger fbshipit-source-id: cfd83e8fb10761d8c9e7f9c020d68c9106a95554
3 years ago
TEST_F(DBTest2, FileTemperatureManifestFixup) {
auto test_fs = std::make_shared<FileTemperatureTestFS>(env_->GetFileSystem());
std::unique_ptr<Env> env(new CompositeEnvWrapper(env_, test_fs));
Options options = CurrentOptions();
options.bottommost_temperature = Temperature::kWarm;
// set dynamic_level to true so the compaction would compact the data to the
// last level directly which will have the last_level_temperature
options.level_compaction_dynamic_level_bytes = true;
Add manifest fix-up utility for file temperatures (#9683) Summary: The goal of this change is to allow changes to the "current" (in FileSystem) file temperatures to feed back into DB metadata, so that they can inform decisions and stats reporting. In part because of modular code factoring, it doesn't seem easy to do this automagically, where opening an SST file and observing current Temperature different from expected would trigger a change in metadata and DB manifest write (essentially giving the deep read path access to the write path). It is also difficult to do this while the DB is open because of the limitations of LogAndApply. This change allows updating file temperature metadata on a closed DB using an experimental utility function UpdateManifestForFilesState() or `ldb update_manifest --update_temperatures`. This should suffice for "migration" scenarios where outside tooling has placed or re-arranged DB files into a (different) tiered configuration without going through RocksDB itself (currently, only compaction can change temperature metadata). Some details: * Refactored and added unit test for `ldb unsafe_remove_sst_file` because of shared functionality * Pulled in autovector.h changes from https://github.com/facebook/rocksdb/issues/9546 to fix SuperVersionContext move constructor (related to an older draft of this change) Possible follow-up work: * Support updating manifest with file checksums, such as when a new checksum function is used and want existing DB metadata updated for it. * It's possible that for some repair scenarios, lighter weight than full repair, we might want to support UpdateManifestForFilesState() to modify critical file details like size or checksum using same algorithm. But let's make sure these are differentiated from modifying file details in ways that don't suspect corruption (or require extreme trust). Pull Request resolved: https://github.com/facebook/rocksdb/pull/9683 Test Plan: unit tests added Reviewed By: jay-zhuang Differential Revision: D34798828 Pulled By: pdillinger fbshipit-source-id: cfd83e8fb10761d8c9e7f9c020d68c9106a95554
3 years ago
options.level0_file_num_compaction_trigger = 2;
options.env = env.get();
std::vector<std::string> cfs = {/*"default",*/ "test1", "test2"};
CreateAndReopenWithCF(cfs, options);
// Needed for later re-opens (weird)
cfs.insert(cfs.begin(), kDefaultColumnFamilyName);
// Generate a bottommost file in all CFs
for (int cf = 0; cf < 3; ++cf) {
ASSERT_OK(Put(cf, "a", "val"));
ASSERT_OK(Put(cf, "c", "val"));
ASSERT_OK(Flush(cf));
ASSERT_OK(Put(cf, "b", "val"));
ASSERT_OK(Put(cf, "d", "val"));
ASSERT_OK(Flush(cf));
}
ASSERT_OK(dbfull()->TEST_WaitForCompact());
// verify
ASSERT_GT(GetSstSizeHelper(Temperature::kWarm), 0);
ASSERT_EQ(GetSstSizeHelper(Temperature::kUnknown), 0);
ASSERT_EQ(GetSstSizeHelper(Temperature::kCold), 0);
ASSERT_EQ(GetSstSizeHelper(Temperature::kHot), 0);
// Generate a non-bottommost file in all CFs
for (int cf = 0; cf < 3; ++cf) {
ASSERT_OK(Put(cf, "e", "val"));
ASSERT_OK(Flush(cf));
}
// re-verify
ASSERT_GT(GetSstSizeHelper(Temperature::kWarm), 0);
// Not supported: ASSERT_GT(GetSstSizeHelper(Temperature::kUnknown), 0);
ASSERT_EQ(GetSstSizeHelper(Temperature::kCold), 0);
ASSERT_EQ(GetSstSizeHelper(Temperature::kHot), 0);
// Now change FS temperature on bottommost file(s) to kCold
std::map<uint64_t, Temperature> current_temps;
test_fs->CopyCurrentSstFileTemperatures(&current_temps);
for (auto e : current_temps) {
if (e.second == Temperature::kWarm) {
test_fs->OverrideSstFileTemperature(e.first, Temperature::kCold);
}
}
// Metadata not yet updated
ASSERT_EQ(Get("a"), "val");
ASSERT_EQ(GetSstSizeHelper(Temperature::kCold), 0);
// Update with Close and UpdateManifestForFilesState, but first save cf
// descriptors
std::vector<ColumnFamilyDescriptor> column_families;
for (size_t i = 0; i < handles_.size(); ++i) {
ColumnFamilyDescriptor cfdescriptor;
// GetDescriptor is not implemented for ROCKSDB_LITE
handles_[i]->GetDescriptor(&cfdescriptor).PermitUncheckedError();
column_families.push_back(cfdescriptor);
}
Close();
experimental::UpdateManifestForFilesStateOptions update_opts;
update_opts.update_temperatures = true;
ASSERT_OK(experimental::UpdateManifestForFilesState(
options, dbname_, column_families, update_opts));
// Re-open and re-verify after update
ReopenWithColumnFamilies(cfs, options);
ASSERT_GT(GetSstSizeHelper(Temperature::kCold), 0);
// Not supported: ASSERT_GT(GetSstSizeHelper(Temperature::kUnknown), 0);
ASSERT_EQ(GetSstSizeHelper(Temperature::kWarm), 0);
ASSERT_EQ(GetSstSizeHelper(Temperature::kHot), 0);
// Change kUnknown to kHot
test_fs->CopyCurrentSstFileTemperatures(&current_temps);
for (auto e : current_temps) {
if (e.second == Temperature::kUnknown) {
test_fs->OverrideSstFileTemperature(e.first, Temperature::kHot);
}
}
// Update with Close and UpdateManifestForFilesState
Close();
ASSERT_OK(experimental::UpdateManifestForFilesState(
options, dbname_, column_families, update_opts));
// Re-open and re-verify after update
ReopenWithColumnFamilies(cfs, options);
ASSERT_GT(GetSstSizeHelper(Temperature::kCold), 0);
ASSERT_EQ(GetSstSizeHelper(Temperature::kUnknown), 0);
ASSERT_EQ(GetSstSizeHelper(Temperature::kWarm), 0);
ASSERT_GT(GetSstSizeHelper(Temperature::kHot), 0);
Close();
}
#endif // ROCKSDB_LITE
// WAL recovery mode is WALRecoveryMode::kPointInTimeRecovery.
TEST_F(DBTest2, PointInTimeRecoveryWithIOErrorWhileReadingWal) {
Options options = CurrentOptions();
DestroyAndReopen(options);
ASSERT_OK(Put("foo", "value0"));
Close();
SyncPoint::GetInstance()->DisableProcessing();
SyncPoint::GetInstance()->ClearAllCallBacks();
bool should_inject_error = false;
SyncPoint::GetInstance()->SetCallBack(
"DBImpl::RecoverLogFiles:BeforeReadWal",
[&](void* /*arg*/) { should_inject_error = true; });
SyncPoint::GetInstance()->SetCallBack(
"LogReader::ReadMore:AfterReadFile", [&](void* arg) {
if (should_inject_error) {
ASSERT_NE(nullptr, arg);
*reinterpret_cast<Status*>(arg) = Status::IOError("Injected IOError");
}
});
SyncPoint::GetInstance()->EnableProcessing();
options.avoid_flush_during_recovery = true;
options.wal_recovery_mode = WALRecoveryMode::kPointInTimeRecovery;
Status s = TryReopen(options);
ASSERT_TRUE(s.IsIOError());
}
TEST_F(DBTest2, PointInTimeRecoveryWithSyncFailureInCFCreation) {
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->LoadDependency(
{{"DBImpl::BackgroundCallFlush:Start:1",
"PointInTimeRecoveryWithSyncFailureInCFCreation:1"},
{"PointInTimeRecoveryWithSyncFailureInCFCreation:2",
"DBImpl::BackgroundCallFlush:Start:2"}});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
CreateColumnFamilies({"test1"}, Options());
ASSERT_OK(Put("foo", "bar"));
// Creating a CF when a flush is going on, log is synced but the
// closed log file is not synced and corrupted.
port::Thread flush_thread([&]() { ASSERT_NOK(Flush()); });
TEST_SYNC_POINT("PointInTimeRecoveryWithSyncFailureInCFCreation:1");
CreateColumnFamilies({"test2"}, Options());
env_->corrupt_in_sync_ = true;
TEST_SYNC_POINT("PointInTimeRecoveryWithSyncFailureInCFCreation:2");
flush_thread.join();
env_->corrupt_in_sync_ = false;
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
// Reopening the DB should not corrupt anything
Options options = CurrentOptions();
options.wal_recovery_mode = WALRecoveryMode::kPointInTimeRecovery;
ReopenWithColumnFamilies({"default", "test1", "test2"}, options);
}
TEST_F(DBTest2, RenameDirectory) {
Options options = CurrentOptions();
DestroyAndReopen(options);
ASSERT_OK(Put("foo", "value0"));
Close();
auto old_dbname = dbname_;
auto new_dbname = dbname_ + "_2";
EXPECT_OK(env_->RenameFile(dbname_, new_dbname));
options.create_if_missing = false;
dbname_ = new_dbname;
ASSERT_OK(TryReopen(options));
ASSERT_EQ("value0", Get("foo"));
Destroy(options);
dbname_ = old_dbname;
}
TEST_F(DBTest2, SstUniqueIdVerifyBackwardCompatible) {
const int kNumSst = 3;
const int kLevel0Trigger = 4;
auto options = CurrentOptions();
options.level0_file_num_compaction_trigger = kLevel0Trigger;
options.statistics = CreateDBStatistics();
Always verify SST unique IDs on SST file open (#10532) Summary: Although we've been tracking SST unique IDs in the DB manifest unconditionally, checking has been opt-in and with an extra pass at DB::Open time. This changes the behavior of `verify_sst_unique_id_in_manifest` to check unique ID against manifest every time an SST file is opened through table cache (normal DB operations), replacing the explicit pass over files at DB::Open time. This change also enables the option by default and removes the "EXPERIMENTAL" designation. One possible criticism is that the option no longer ensures the integrity of a DB at Open time. This is far from an all-or-nothing issue. Verifying the IDs of all SST files hardly ensures all the data in the DB is readable. (VerifyChecksum is supposed to do that.) Also, with max_open_files=-1 (default, extremely common), all SST files are opened at DB::Open time anyway. Implementation details: * `VerifySstUniqueIdInManifest()` functions are the extra/explicit pass that is now removed. * Unit tests that manipulate/corrupt table properties have to opt out of this check, because that corrupts the "actual" unique id. (And even for testing we don't currently have a mechanism to set "no unique id" in the in-memory file metadata for new files.) * A lot of other unit test churn relates to (a) default checking on, and (b) checking on SST open even without DB::Open (e.g. on flush) * Use `FileMetaData` for more `TableCache` operations (in place of `FileDescriptor`) so that we have access to the unique_id whenever we might need to open an SST file. **There is the possibility of performance impact because we can no longer use the more localized `fd` part of an `FdWithKeyRange` but instead follow the `file_metadata` pointer. However, this change (possible regression) is only done for `GetMemoryUsageByTableReaders`.** * Removed a completely unnecessary constructor overload of `TableReaderOptions` Possible follow-up: * Verification only happens when opening through table cache. Are there more places where this should happen? * Improve error message when there is a file size mismatch vs. manifest (FIXME added in the appropriate place). * I'm not sure there's a justification for `FileDescriptor` to be distinct from `FileMetaData`. * I'm skeptical that `FdWithKeyRange` really still makes sense for optimizing some data locality by duplicating some data in memory, but I could be wrong. * An unnecessary overload of NewTableReader was recently added, in the public API nonetheless (though unusable there). It should be cleaned up to put most things under `TableReaderOptions`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10532 Test Plan: updated unit tests Performance test showing no significant difference (just noise I think): `./db_bench -benchmarks=readwhilewriting[-X10] -num=3000000 -disable_wal=1 -bloom_bits=8 -write_buffer_size=1000000 -target_file_size_base=1000000` Before: readwhilewriting [AVG 10 runs] : 68702 (± 6932) ops/sec After: readwhilewriting [AVG 10 runs] : 68239 (± 7198) ops/sec Reviewed By: jay-zhuang Differential Revision: D38765551 Pulled By: pdillinger fbshipit-source-id: a827a708155f12344ab2a5c16e7701c7636da4c2
2 years ago
// Skip for now
options.verify_sst_unique_id_in_manifest = false;
Reopen(options);
std::atomic_int skipped = 0;
Always verify SST unique IDs on SST file open (#10532) Summary: Although we've been tracking SST unique IDs in the DB manifest unconditionally, checking has been opt-in and with an extra pass at DB::Open time. This changes the behavior of `verify_sst_unique_id_in_manifest` to check unique ID against manifest every time an SST file is opened through table cache (normal DB operations), replacing the explicit pass over files at DB::Open time. This change also enables the option by default and removes the "EXPERIMENTAL" designation. One possible criticism is that the option no longer ensures the integrity of a DB at Open time. This is far from an all-or-nothing issue. Verifying the IDs of all SST files hardly ensures all the data in the DB is readable. (VerifyChecksum is supposed to do that.) Also, with max_open_files=-1 (default, extremely common), all SST files are opened at DB::Open time anyway. Implementation details: * `VerifySstUniqueIdInManifest()` functions are the extra/explicit pass that is now removed. * Unit tests that manipulate/corrupt table properties have to opt out of this check, because that corrupts the "actual" unique id. (And even for testing we don't currently have a mechanism to set "no unique id" in the in-memory file metadata for new files.) * A lot of other unit test churn relates to (a) default checking on, and (b) checking on SST open even without DB::Open (e.g. on flush) * Use `FileMetaData` for more `TableCache` operations (in place of `FileDescriptor`) so that we have access to the unique_id whenever we might need to open an SST file. **There is the possibility of performance impact because we can no longer use the more localized `fd` part of an `FdWithKeyRange` but instead follow the `file_metadata` pointer. However, this change (possible regression) is only done for `GetMemoryUsageByTableReaders`.** * Removed a completely unnecessary constructor overload of `TableReaderOptions` Possible follow-up: * Verification only happens when opening through table cache. Are there more places where this should happen? * Improve error message when there is a file size mismatch vs. manifest (FIXME added in the appropriate place). * I'm not sure there's a justification for `FileDescriptor` to be distinct from `FileMetaData`. * I'm skeptical that `FdWithKeyRange` really still makes sense for optimizing some data locality by duplicating some data in memory, but I could be wrong. * An unnecessary overload of NewTableReader was recently added, in the public API nonetheless (though unusable there). It should be cleaned up to put most things under `TableReaderOptions`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10532 Test Plan: updated unit tests Performance test showing no significant difference (just noise I think): `./db_bench -benchmarks=readwhilewriting[-X10] -num=3000000 -disable_wal=1 -bloom_bits=8 -write_buffer_size=1000000 -target_file_size_base=1000000` Before: readwhilewriting [AVG 10 runs] : 68702 (± 6932) ops/sec After: readwhilewriting [AVG 10 runs] : 68239 (± 7198) ops/sec Reviewed By: jay-zhuang Differential Revision: D38765551 Pulled By: pdillinger fbshipit-source-id: a827a708155f12344ab2a5c16e7701c7636da4c2
2 years ago
std::atomic_int passed = 0;
SyncPoint::GetInstance()->SetCallBack(
"BlockBasedTable::Open::SkippedVerifyUniqueId",
[&](void* /*arg*/) { skipped++; });
SyncPoint::GetInstance()->SetCallBack(
"BlockBasedTable::Open::PassedVerifyUniqueId",
[&](void* /*arg*/) { passed++; });
SyncPoint::GetInstance()->EnableProcessing();
// generate a few SSTs
for (int i = 0; i < kNumSst; i++) {
for (int j = 0; j < 100; j++) {
ASSERT_OK(Put(Key(i * 10 + j), "value"));
}
ASSERT_OK(Flush());
}
Always verify SST unique IDs on SST file open (#10532) Summary: Although we've been tracking SST unique IDs in the DB manifest unconditionally, checking has been opt-in and with an extra pass at DB::Open time. This changes the behavior of `verify_sst_unique_id_in_manifest` to check unique ID against manifest every time an SST file is opened through table cache (normal DB operations), replacing the explicit pass over files at DB::Open time. This change also enables the option by default and removes the "EXPERIMENTAL" designation. One possible criticism is that the option no longer ensures the integrity of a DB at Open time. This is far from an all-or-nothing issue. Verifying the IDs of all SST files hardly ensures all the data in the DB is readable. (VerifyChecksum is supposed to do that.) Also, with max_open_files=-1 (default, extremely common), all SST files are opened at DB::Open time anyway. Implementation details: * `VerifySstUniqueIdInManifest()` functions are the extra/explicit pass that is now removed. * Unit tests that manipulate/corrupt table properties have to opt out of this check, because that corrupts the "actual" unique id. (And even for testing we don't currently have a mechanism to set "no unique id" in the in-memory file metadata for new files.) * A lot of other unit test churn relates to (a) default checking on, and (b) checking on SST open even without DB::Open (e.g. on flush) * Use `FileMetaData` for more `TableCache` operations (in place of `FileDescriptor`) so that we have access to the unique_id whenever we might need to open an SST file. **There is the possibility of performance impact because we can no longer use the more localized `fd` part of an `FdWithKeyRange` but instead follow the `file_metadata` pointer. However, this change (possible regression) is only done for `GetMemoryUsageByTableReaders`.** * Removed a completely unnecessary constructor overload of `TableReaderOptions` Possible follow-up: * Verification only happens when opening through table cache. Are there more places where this should happen? * Improve error message when there is a file size mismatch vs. manifest (FIXME added in the appropriate place). * I'm not sure there's a justification for `FileDescriptor` to be distinct from `FileMetaData`. * I'm skeptical that `FdWithKeyRange` really still makes sense for optimizing some data locality by duplicating some data in memory, but I could be wrong. * An unnecessary overload of NewTableReader was recently added, in the public API nonetheless (though unusable there). It should be cleaned up to put most things under `TableReaderOptions`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10532 Test Plan: updated unit tests Performance test showing no significant difference (just noise I think): `./db_bench -benchmarks=readwhilewriting[-X10] -num=3000000 -disable_wal=1 -bloom_bits=8 -write_buffer_size=1000000 -target_file_size_base=1000000` Before: readwhilewriting [AVG 10 runs] : 68702 (± 6932) ops/sec After: readwhilewriting [AVG 10 runs] : 68239 (± 7198) ops/sec Reviewed By: jay-zhuang Differential Revision: D38765551 Pulled By: pdillinger fbshipit-source-id: a827a708155f12344ab2a5c16e7701c7636da4c2
2 years ago
// Verification has been skipped on files so far
EXPECT_EQ(skipped, kNumSst);
EXPECT_EQ(passed, 0);
Always verify SST unique IDs on SST file open (#10532) Summary: Although we've been tracking SST unique IDs in the DB manifest unconditionally, checking has been opt-in and with an extra pass at DB::Open time. This changes the behavior of `verify_sst_unique_id_in_manifest` to check unique ID against manifest every time an SST file is opened through table cache (normal DB operations), replacing the explicit pass over files at DB::Open time. This change also enables the option by default and removes the "EXPERIMENTAL" designation. One possible criticism is that the option no longer ensures the integrity of a DB at Open time. This is far from an all-or-nothing issue. Verifying the IDs of all SST files hardly ensures all the data in the DB is readable. (VerifyChecksum is supposed to do that.) Also, with max_open_files=-1 (default, extremely common), all SST files are opened at DB::Open time anyway. Implementation details: * `VerifySstUniqueIdInManifest()` functions are the extra/explicit pass that is now removed. * Unit tests that manipulate/corrupt table properties have to opt out of this check, because that corrupts the "actual" unique id. (And even for testing we don't currently have a mechanism to set "no unique id" in the in-memory file metadata for new files.) * A lot of other unit test churn relates to (a) default checking on, and (b) checking on SST open even without DB::Open (e.g. on flush) * Use `FileMetaData` for more `TableCache` operations (in place of `FileDescriptor`) so that we have access to the unique_id whenever we might need to open an SST file. **There is the possibility of performance impact because we can no longer use the more localized `fd` part of an `FdWithKeyRange` but instead follow the `file_metadata` pointer. However, this change (possible regression) is only done for `GetMemoryUsageByTableReaders`.** * Removed a completely unnecessary constructor overload of `TableReaderOptions` Possible follow-up: * Verification only happens when opening through table cache. Are there more places where this should happen? * Improve error message when there is a file size mismatch vs. manifest (FIXME added in the appropriate place). * I'm not sure there's a justification for `FileDescriptor` to be distinct from `FileMetaData`. * I'm skeptical that `FdWithKeyRange` really still makes sense for optimizing some data locality by duplicating some data in memory, but I could be wrong. * An unnecessary overload of NewTableReader was recently added, in the public API nonetheless (though unusable there). It should be cleaned up to put most things under `TableReaderOptions`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10532 Test Plan: updated unit tests Performance test showing no significant difference (just noise I think): `./db_bench -benchmarks=readwhilewriting[-X10] -num=3000000 -disable_wal=1 -bloom_bits=8 -write_buffer_size=1000000 -target_file_size_base=1000000` Before: readwhilewriting [AVG 10 runs] : 68702 (± 6932) ops/sec After: readwhilewriting [AVG 10 runs] : 68239 (± 7198) ops/sec Reviewed By: jay-zhuang Differential Revision: D38765551 Pulled By: pdillinger fbshipit-source-id: a827a708155f12344ab2a5c16e7701c7636da4c2
2 years ago
// Reopen with verification
options.verify_sst_unique_id_in_manifest = true;
Always verify SST unique IDs on SST file open (#10532) Summary: Although we've been tracking SST unique IDs in the DB manifest unconditionally, checking has been opt-in and with an extra pass at DB::Open time. This changes the behavior of `verify_sst_unique_id_in_manifest` to check unique ID against manifest every time an SST file is opened through table cache (normal DB operations), replacing the explicit pass over files at DB::Open time. This change also enables the option by default and removes the "EXPERIMENTAL" designation. One possible criticism is that the option no longer ensures the integrity of a DB at Open time. This is far from an all-or-nothing issue. Verifying the IDs of all SST files hardly ensures all the data in the DB is readable. (VerifyChecksum is supposed to do that.) Also, with max_open_files=-1 (default, extremely common), all SST files are opened at DB::Open time anyway. Implementation details: * `VerifySstUniqueIdInManifest()` functions are the extra/explicit pass that is now removed. * Unit tests that manipulate/corrupt table properties have to opt out of this check, because that corrupts the "actual" unique id. (And even for testing we don't currently have a mechanism to set "no unique id" in the in-memory file metadata for new files.) * A lot of other unit test churn relates to (a) default checking on, and (b) checking on SST open even without DB::Open (e.g. on flush) * Use `FileMetaData` for more `TableCache` operations (in place of `FileDescriptor`) so that we have access to the unique_id whenever we might need to open an SST file. **There is the possibility of performance impact because we can no longer use the more localized `fd` part of an `FdWithKeyRange` but instead follow the `file_metadata` pointer. However, this change (possible regression) is only done for `GetMemoryUsageByTableReaders`.** * Removed a completely unnecessary constructor overload of `TableReaderOptions` Possible follow-up: * Verification only happens when opening through table cache. Are there more places where this should happen? * Improve error message when there is a file size mismatch vs. manifest (FIXME added in the appropriate place). * I'm not sure there's a justification for `FileDescriptor` to be distinct from `FileMetaData`. * I'm skeptical that `FdWithKeyRange` really still makes sense for optimizing some data locality by duplicating some data in memory, but I could be wrong. * An unnecessary overload of NewTableReader was recently added, in the public API nonetheless (though unusable there). It should be cleaned up to put most things under `TableReaderOptions`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10532 Test Plan: updated unit tests Performance test showing no significant difference (just noise I think): `./db_bench -benchmarks=readwhilewriting[-X10] -num=3000000 -disable_wal=1 -bloom_bits=8 -write_buffer_size=1000000 -target_file_size_base=1000000` Before: readwhilewriting [AVG 10 runs] : 68702 (± 6932) ops/sec After: readwhilewriting [AVG 10 runs] : 68239 (± 7198) ops/sec Reviewed By: jay-zhuang Differential Revision: D38765551 Pulled By: pdillinger fbshipit-source-id: a827a708155f12344ab2a5c16e7701c7636da4c2
2 years ago
skipped = 0;
passed = 0;
Reopen(options);
Always verify SST unique IDs on SST file open (#10532) Summary: Although we've been tracking SST unique IDs in the DB manifest unconditionally, checking has been opt-in and with an extra pass at DB::Open time. This changes the behavior of `verify_sst_unique_id_in_manifest` to check unique ID against manifest every time an SST file is opened through table cache (normal DB operations), replacing the explicit pass over files at DB::Open time. This change also enables the option by default and removes the "EXPERIMENTAL" designation. One possible criticism is that the option no longer ensures the integrity of a DB at Open time. This is far from an all-or-nothing issue. Verifying the IDs of all SST files hardly ensures all the data in the DB is readable. (VerifyChecksum is supposed to do that.) Also, with max_open_files=-1 (default, extremely common), all SST files are opened at DB::Open time anyway. Implementation details: * `VerifySstUniqueIdInManifest()` functions are the extra/explicit pass that is now removed. * Unit tests that manipulate/corrupt table properties have to opt out of this check, because that corrupts the "actual" unique id. (And even for testing we don't currently have a mechanism to set "no unique id" in the in-memory file metadata for new files.) * A lot of other unit test churn relates to (a) default checking on, and (b) checking on SST open even without DB::Open (e.g. on flush) * Use `FileMetaData` for more `TableCache` operations (in place of `FileDescriptor`) so that we have access to the unique_id whenever we might need to open an SST file. **There is the possibility of performance impact because we can no longer use the more localized `fd` part of an `FdWithKeyRange` but instead follow the `file_metadata` pointer. However, this change (possible regression) is only done for `GetMemoryUsageByTableReaders`.** * Removed a completely unnecessary constructor overload of `TableReaderOptions` Possible follow-up: * Verification only happens when opening through table cache. Are there more places where this should happen? * Improve error message when there is a file size mismatch vs. manifest (FIXME added in the appropriate place). * I'm not sure there's a justification for `FileDescriptor` to be distinct from `FileMetaData`. * I'm skeptical that `FdWithKeyRange` really still makes sense for optimizing some data locality by duplicating some data in memory, but I could be wrong. * An unnecessary overload of NewTableReader was recently added, in the public API nonetheless (though unusable there). It should be cleaned up to put most things under `TableReaderOptions`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10532 Test Plan: updated unit tests Performance test showing no significant difference (just noise I think): `./db_bench -benchmarks=readwhilewriting[-X10] -num=3000000 -disable_wal=1 -bloom_bits=8 -write_buffer_size=1000000 -target_file_size_base=1000000` Before: readwhilewriting [AVG 10 runs] : 68702 (± 6932) ops/sec After: readwhilewriting [AVG 10 runs] : 68239 (± 7198) ops/sec Reviewed By: jay-zhuang Differential Revision: D38765551 Pulled By: pdillinger fbshipit-source-id: a827a708155f12344ab2a5c16e7701c7636da4c2
2 years ago
EXPECT_EQ(skipped, 0);
EXPECT_EQ(passed, kNumSst);
// Now simulate no unique id in manifest for next file
// NOTE: this only works for loading manifest from disk,
// not in-memory manifest, so we need to re-open below.
SyncPoint::GetInstance()->SetCallBack(
"VersionEdit::EncodeTo:UniqueId", [&](void* arg) {
auto unique_id = static_cast<UniqueId64x2*>(arg);
// remove id before writing it to manifest
(*unique_id)[0] = 0;
(*unique_id)[1] = 0;
});
// test compaction generated Sst
for (int i = kNumSst; i < kLevel0Trigger; i++) {
for (int j = 0; j < 100; j++) {
ASSERT_OK(Put(Key(i * 10 + j), "value"));
}
ASSERT_OK(Flush());
}
ASSERT_OK(dbfull()->TEST_WaitForCompact());
#ifndef ROCKSDB_LITE
ASSERT_EQ("0,1", FilesPerLevel(0));
#endif // ROCKSDB_LITE
Always verify SST unique IDs on SST file open (#10532) Summary: Although we've been tracking SST unique IDs in the DB manifest unconditionally, checking has been opt-in and with an extra pass at DB::Open time. This changes the behavior of `verify_sst_unique_id_in_manifest` to check unique ID against manifest every time an SST file is opened through table cache (normal DB operations), replacing the explicit pass over files at DB::Open time. This change also enables the option by default and removes the "EXPERIMENTAL" designation. One possible criticism is that the option no longer ensures the integrity of a DB at Open time. This is far from an all-or-nothing issue. Verifying the IDs of all SST files hardly ensures all the data in the DB is readable. (VerifyChecksum is supposed to do that.) Also, with max_open_files=-1 (default, extremely common), all SST files are opened at DB::Open time anyway. Implementation details: * `VerifySstUniqueIdInManifest()` functions are the extra/explicit pass that is now removed. * Unit tests that manipulate/corrupt table properties have to opt out of this check, because that corrupts the "actual" unique id. (And even for testing we don't currently have a mechanism to set "no unique id" in the in-memory file metadata for new files.) * A lot of other unit test churn relates to (a) default checking on, and (b) checking on SST open even without DB::Open (e.g. on flush) * Use `FileMetaData` for more `TableCache` operations (in place of `FileDescriptor`) so that we have access to the unique_id whenever we might need to open an SST file. **There is the possibility of performance impact because we can no longer use the more localized `fd` part of an `FdWithKeyRange` but instead follow the `file_metadata` pointer. However, this change (possible regression) is only done for `GetMemoryUsageByTableReaders`.** * Removed a completely unnecessary constructor overload of `TableReaderOptions` Possible follow-up: * Verification only happens when opening through table cache. Are there more places where this should happen? * Improve error message when there is a file size mismatch vs. manifest (FIXME added in the appropriate place). * I'm not sure there's a justification for `FileDescriptor` to be distinct from `FileMetaData`. * I'm skeptical that `FdWithKeyRange` really still makes sense for optimizing some data locality by duplicating some data in memory, but I could be wrong. * An unnecessary overload of NewTableReader was recently added, in the public API nonetheless (though unusable there). It should be cleaned up to put most things under `TableReaderOptions`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10532 Test Plan: updated unit tests Performance test showing no significant difference (just noise I think): `./db_bench -benchmarks=readwhilewriting[-X10] -num=3000000 -disable_wal=1 -bloom_bits=8 -write_buffer_size=1000000 -target_file_size_base=1000000` Before: readwhilewriting [AVG 10 runs] : 68702 (± 6932) ops/sec After: readwhilewriting [AVG 10 runs] : 68239 (± 7198) ops/sec Reviewed By: jay-zhuang Differential Revision: D38765551 Pulled By: pdillinger fbshipit-source-id: a827a708155f12344ab2a5c16e7701c7636da4c2
2 years ago
// Reopen (with verification)
ASSERT_TRUE(options.verify_sst_unique_id_in_manifest);
skipped = 0;
Always verify SST unique IDs on SST file open (#10532) Summary: Although we've been tracking SST unique IDs in the DB manifest unconditionally, checking has been opt-in and with an extra pass at DB::Open time. This changes the behavior of `verify_sst_unique_id_in_manifest` to check unique ID against manifest every time an SST file is opened through table cache (normal DB operations), replacing the explicit pass over files at DB::Open time. This change also enables the option by default and removes the "EXPERIMENTAL" designation. One possible criticism is that the option no longer ensures the integrity of a DB at Open time. This is far from an all-or-nothing issue. Verifying the IDs of all SST files hardly ensures all the data in the DB is readable. (VerifyChecksum is supposed to do that.) Also, with max_open_files=-1 (default, extremely common), all SST files are opened at DB::Open time anyway. Implementation details: * `VerifySstUniqueIdInManifest()` functions are the extra/explicit pass that is now removed. * Unit tests that manipulate/corrupt table properties have to opt out of this check, because that corrupts the "actual" unique id. (And even for testing we don't currently have a mechanism to set "no unique id" in the in-memory file metadata for new files.) * A lot of other unit test churn relates to (a) default checking on, and (b) checking on SST open even without DB::Open (e.g. on flush) * Use `FileMetaData` for more `TableCache` operations (in place of `FileDescriptor`) so that we have access to the unique_id whenever we might need to open an SST file. **There is the possibility of performance impact because we can no longer use the more localized `fd` part of an `FdWithKeyRange` but instead follow the `file_metadata` pointer. However, this change (possible regression) is only done for `GetMemoryUsageByTableReaders`.** * Removed a completely unnecessary constructor overload of `TableReaderOptions` Possible follow-up: * Verification only happens when opening through table cache. Are there more places where this should happen? * Improve error message when there is a file size mismatch vs. manifest (FIXME added in the appropriate place). * I'm not sure there's a justification for `FileDescriptor` to be distinct from `FileMetaData`. * I'm skeptical that `FdWithKeyRange` really still makes sense for optimizing some data locality by duplicating some data in memory, but I could be wrong. * An unnecessary overload of NewTableReader was recently added, in the public API nonetheless (though unusable there). It should be cleaned up to put most things under `TableReaderOptions`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10532 Test Plan: updated unit tests Performance test showing no significant difference (just noise I think): `./db_bench -benchmarks=readwhilewriting[-X10] -num=3000000 -disable_wal=1 -bloom_bits=8 -write_buffer_size=1000000 -target_file_size_base=1000000` Before: readwhilewriting [AVG 10 runs] : 68702 (± 6932) ops/sec After: readwhilewriting [AVG 10 runs] : 68239 (± 7198) ops/sec Reviewed By: jay-zhuang Differential Revision: D38765551 Pulled By: pdillinger fbshipit-source-id: a827a708155f12344ab2a5c16e7701c7636da4c2
2 years ago
passed = 0;
Reopen(options);
Always verify SST unique IDs on SST file open (#10532) Summary: Although we've been tracking SST unique IDs in the DB manifest unconditionally, checking has been opt-in and with an extra pass at DB::Open time. This changes the behavior of `verify_sst_unique_id_in_manifest` to check unique ID against manifest every time an SST file is opened through table cache (normal DB operations), replacing the explicit pass over files at DB::Open time. This change also enables the option by default and removes the "EXPERIMENTAL" designation. One possible criticism is that the option no longer ensures the integrity of a DB at Open time. This is far from an all-or-nothing issue. Verifying the IDs of all SST files hardly ensures all the data in the DB is readable. (VerifyChecksum is supposed to do that.) Also, with max_open_files=-1 (default, extremely common), all SST files are opened at DB::Open time anyway. Implementation details: * `VerifySstUniqueIdInManifest()` functions are the extra/explicit pass that is now removed. * Unit tests that manipulate/corrupt table properties have to opt out of this check, because that corrupts the "actual" unique id. (And even for testing we don't currently have a mechanism to set "no unique id" in the in-memory file metadata for new files.) * A lot of other unit test churn relates to (a) default checking on, and (b) checking on SST open even without DB::Open (e.g. on flush) * Use `FileMetaData` for more `TableCache` operations (in place of `FileDescriptor`) so that we have access to the unique_id whenever we might need to open an SST file. **There is the possibility of performance impact because we can no longer use the more localized `fd` part of an `FdWithKeyRange` but instead follow the `file_metadata` pointer. However, this change (possible regression) is only done for `GetMemoryUsageByTableReaders`.** * Removed a completely unnecessary constructor overload of `TableReaderOptions` Possible follow-up: * Verification only happens when opening through table cache. Are there more places where this should happen? * Improve error message when there is a file size mismatch vs. manifest (FIXME added in the appropriate place). * I'm not sure there's a justification for `FileDescriptor` to be distinct from `FileMetaData`. * I'm skeptical that `FdWithKeyRange` really still makes sense for optimizing some data locality by duplicating some data in memory, but I could be wrong. * An unnecessary overload of NewTableReader was recently added, in the public API nonetheless (though unusable there). It should be cleaned up to put most things under `TableReaderOptions`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10532 Test Plan: updated unit tests Performance test showing no significant difference (just noise I think): `./db_bench -benchmarks=readwhilewriting[-X10] -num=3000000 -disable_wal=1 -bloom_bits=8 -write_buffer_size=1000000 -target_file_size_base=1000000` Before: readwhilewriting [AVG 10 runs] : 68702 (± 6932) ops/sec After: readwhilewriting [AVG 10 runs] : 68239 (± 7198) ops/sec Reviewed By: jay-zhuang Differential Revision: D38765551 Pulled By: pdillinger fbshipit-source-id: a827a708155f12344ab2a5c16e7701c7636da4c2
2 years ago
EXPECT_EQ(skipped, 1);
EXPECT_EQ(passed, 0);
}
TEST_F(DBTest2, SstUniqueIdVerify) {
const int kNumSst = 3;
const int kLevel0Trigger = 4;
auto options = CurrentOptions();
options.level0_file_num_compaction_trigger = kLevel0Trigger;
Always verify SST unique IDs on SST file open (#10532) Summary: Although we've been tracking SST unique IDs in the DB manifest unconditionally, checking has been opt-in and with an extra pass at DB::Open time. This changes the behavior of `verify_sst_unique_id_in_manifest` to check unique ID against manifest every time an SST file is opened through table cache (normal DB operations), replacing the explicit pass over files at DB::Open time. This change also enables the option by default and removes the "EXPERIMENTAL" designation. One possible criticism is that the option no longer ensures the integrity of a DB at Open time. This is far from an all-or-nothing issue. Verifying the IDs of all SST files hardly ensures all the data in the DB is readable. (VerifyChecksum is supposed to do that.) Also, with max_open_files=-1 (default, extremely common), all SST files are opened at DB::Open time anyway. Implementation details: * `VerifySstUniqueIdInManifest()` functions are the extra/explicit pass that is now removed. * Unit tests that manipulate/corrupt table properties have to opt out of this check, because that corrupts the "actual" unique id. (And even for testing we don't currently have a mechanism to set "no unique id" in the in-memory file metadata for new files.) * A lot of other unit test churn relates to (a) default checking on, and (b) checking on SST open even without DB::Open (e.g. on flush) * Use `FileMetaData` for more `TableCache` operations (in place of `FileDescriptor`) so that we have access to the unique_id whenever we might need to open an SST file. **There is the possibility of performance impact because we can no longer use the more localized `fd` part of an `FdWithKeyRange` but instead follow the `file_metadata` pointer. However, this change (possible regression) is only done for `GetMemoryUsageByTableReaders`.** * Removed a completely unnecessary constructor overload of `TableReaderOptions` Possible follow-up: * Verification only happens when opening through table cache. Are there more places where this should happen? * Improve error message when there is a file size mismatch vs. manifest (FIXME added in the appropriate place). * I'm not sure there's a justification for `FileDescriptor` to be distinct from `FileMetaData`. * I'm skeptical that `FdWithKeyRange` really still makes sense for optimizing some data locality by duplicating some data in memory, but I could be wrong. * An unnecessary overload of NewTableReader was recently added, in the public API nonetheless (though unusable there). It should be cleaned up to put most things under `TableReaderOptions`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10532 Test Plan: updated unit tests Performance test showing no significant difference (just noise I think): `./db_bench -benchmarks=readwhilewriting[-X10] -num=3000000 -disable_wal=1 -bloom_bits=8 -write_buffer_size=1000000 -target_file_size_base=1000000` Before: readwhilewriting [AVG 10 runs] : 68702 (± 6932) ops/sec After: readwhilewriting [AVG 10 runs] : 68239 (± 7198) ops/sec Reviewed By: jay-zhuang Differential Revision: D38765551 Pulled By: pdillinger fbshipit-source-id: a827a708155f12344ab2a5c16e7701c7636da4c2
2 years ago
// Allow mismatch for now
options.verify_sst_unique_id_in_manifest = false;
Reopen(options);
SyncPoint::GetInstance()->SetCallBack(
"PropertyBlockBuilder::AddTableProperty:Start", [&](void* props_vs) {
auto props = static_cast<TableProperties*>(props_vs);
Always verify SST unique IDs on SST file open (#10532) Summary: Although we've been tracking SST unique IDs in the DB manifest unconditionally, checking has been opt-in and with an extra pass at DB::Open time. This changes the behavior of `verify_sst_unique_id_in_manifest` to check unique ID against manifest every time an SST file is opened through table cache (normal DB operations), replacing the explicit pass over files at DB::Open time. This change also enables the option by default and removes the "EXPERIMENTAL" designation. One possible criticism is that the option no longer ensures the integrity of a DB at Open time. This is far from an all-or-nothing issue. Verifying the IDs of all SST files hardly ensures all the data in the DB is readable. (VerifyChecksum is supposed to do that.) Also, with max_open_files=-1 (default, extremely common), all SST files are opened at DB::Open time anyway. Implementation details: * `VerifySstUniqueIdInManifest()` functions are the extra/explicit pass that is now removed. * Unit tests that manipulate/corrupt table properties have to opt out of this check, because that corrupts the "actual" unique id. (And even for testing we don't currently have a mechanism to set "no unique id" in the in-memory file metadata for new files.) * A lot of other unit test churn relates to (a) default checking on, and (b) checking on SST open even without DB::Open (e.g. on flush) * Use `FileMetaData` for more `TableCache` operations (in place of `FileDescriptor`) so that we have access to the unique_id whenever we might need to open an SST file. **There is the possibility of performance impact because we can no longer use the more localized `fd` part of an `FdWithKeyRange` but instead follow the `file_metadata` pointer. However, this change (possible regression) is only done for `GetMemoryUsageByTableReaders`.** * Removed a completely unnecessary constructor overload of `TableReaderOptions` Possible follow-up: * Verification only happens when opening through table cache. Are there more places where this should happen? * Improve error message when there is a file size mismatch vs. manifest (FIXME added in the appropriate place). * I'm not sure there's a justification for `FileDescriptor` to be distinct from `FileMetaData`. * I'm skeptical that `FdWithKeyRange` really still makes sense for optimizing some data locality by duplicating some data in memory, but I could be wrong. * An unnecessary overload of NewTableReader was recently added, in the public API nonetheless (though unusable there). It should be cleaned up to put most things under `TableReaderOptions`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10532 Test Plan: updated unit tests Performance test showing no significant difference (just noise I think): `./db_bench -benchmarks=readwhilewriting[-X10] -num=3000000 -disable_wal=1 -bloom_bits=8 -write_buffer_size=1000000 -target_file_size_base=1000000` Before: readwhilewriting [AVG 10 runs] : 68702 (± 6932) ops/sec After: readwhilewriting [AVG 10 runs] : 68239 (± 7198) ops/sec Reviewed By: jay-zhuang Differential Revision: D38765551 Pulled By: pdillinger fbshipit-source-id: a827a708155f12344ab2a5c16e7701c7636da4c2
2 years ago
// update table property session_id to a different one, which
// changes unique ID
props->db_session_id = DBImpl::GenerateDbSessionId(nullptr);
});
SyncPoint::GetInstance()->EnableProcessing();
// generate a few SSTs
for (int i = 0; i < kNumSst; i++) {
for (int j = 0; j < 100; j++) {
ASSERT_OK(Put(Key(i * 10 + j), "value"));
}
ASSERT_OK(Flush());
}
// Reopen with verification should report corruption
options.verify_sst_unique_id_in_manifest = true;
auto s = TryReopen(options);
ASSERT_TRUE(s.IsCorruption());
// Reopen without verification should be fine
options.verify_sst_unique_id_in_manifest = false;
Reopen(options);
// test compaction generated Sst
for (int i = kNumSst; i < kLevel0Trigger; i++) {
for (int j = 0; j < 100; j++) {
ASSERT_OK(Put(Key(i * 10 + j), "value"));
}
ASSERT_OK(Flush());
}
ASSERT_OK(dbfull()->TEST_WaitForCompact());
#ifndef ROCKSDB_LITE
ASSERT_EQ("0,1", FilesPerLevel(0));
#endif // ROCKSDB_LITE
// Reopen with verification should fail
options.verify_sst_unique_id_in_manifest = true;
s = TryReopen(options);
ASSERT_TRUE(s.IsCorruption());
}
TEST_F(DBTest2, SstUniqueIdVerifyMultiCFs) {
const int kNumSst = 3;
const int kLevel0Trigger = 4;
auto options = CurrentOptions();
options.level0_file_num_compaction_trigger = kLevel0Trigger;
Always verify SST unique IDs on SST file open (#10532) Summary: Although we've been tracking SST unique IDs in the DB manifest unconditionally, checking has been opt-in and with an extra pass at DB::Open time. This changes the behavior of `verify_sst_unique_id_in_manifest` to check unique ID against manifest every time an SST file is opened through table cache (normal DB operations), replacing the explicit pass over files at DB::Open time. This change also enables the option by default and removes the "EXPERIMENTAL" designation. One possible criticism is that the option no longer ensures the integrity of a DB at Open time. This is far from an all-or-nothing issue. Verifying the IDs of all SST files hardly ensures all the data in the DB is readable. (VerifyChecksum is supposed to do that.) Also, with max_open_files=-1 (default, extremely common), all SST files are opened at DB::Open time anyway. Implementation details: * `VerifySstUniqueIdInManifest()` functions are the extra/explicit pass that is now removed. * Unit tests that manipulate/corrupt table properties have to opt out of this check, because that corrupts the "actual" unique id. (And even for testing we don't currently have a mechanism to set "no unique id" in the in-memory file metadata for new files.) * A lot of other unit test churn relates to (a) default checking on, and (b) checking on SST open even without DB::Open (e.g. on flush) * Use `FileMetaData` for more `TableCache` operations (in place of `FileDescriptor`) so that we have access to the unique_id whenever we might need to open an SST file. **There is the possibility of performance impact because we can no longer use the more localized `fd` part of an `FdWithKeyRange` but instead follow the `file_metadata` pointer. However, this change (possible regression) is only done for `GetMemoryUsageByTableReaders`.** * Removed a completely unnecessary constructor overload of `TableReaderOptions` Possible follow-up: * Verification only happens when opening through table cache. Are there more places where this should happen? * Improve error message when there is a file size mismatch vs. manifest (FIXME added in the appropriate place). * I'm not sure there's a justification for `FileDescriptor` to be distinct from `FileMetaData`. * I'm skeptical that `FdWithKeyRange` really still makes sense for optimizing some data locality by duplicating some data in memory, but I could be wrong. * An unnecessary overload of NewTableReader was recently added, in the public API nonetheless (though unusable there). It should be cleaned up to put most things under `TableReaderOptions`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10532 Test Plan: updated unit tests Performance test showing no significant difference (just noise I think): `./db_bench -benchmarks=readwhilewriting[-X10] -num=3000000 -disable_wal=1 -bloom_bits=8 -write_buffer_size=1000000 -target_file_size_base=1000000` Before: readwhilewriting [AVG 10 runs] : 68702 (± 6932) ops/sec After: readwhilewriting [AVG 10 runs] : 68239 (± 7198) ops/sec Reviewed By: jay-zhuang Differential Revision: D38765551 Pulled By: pdillinger fbshipit-source-id: a827a708155f12344ab2a5c16e7701c7636da4c2
2 years ago
// Allow mismatch for now
options.verify_sst_unique_id_in_manifest = false;
CreateAndReopenWithCF({"one", "two"}, options);
// generate good SSTs
for (int cf_num : {0, 2}) {
for (int i = 0; i < kNumSst; i++) {
for (int j = 0; j < 100; j++) {
ASSERT_OK(Put(cf_num, Key(i * 10 + j), "value"));
}
ASSERT_OK(Flush(cf_num));
}
}
// generate SSTs with bad unique id
SyncPoint::GetInstance()->SetCallBack(
"PropertyBlockBuilder::AddTableProperty:Start", [&](void* props_vs) {
auto props = static_cast<TableProperties*>(props_vs);
// update table property session_id to a different one
props->db_session_id = DBImpl::GenerateDbSessionId(nullptr);
});
SyncPoint::GetInstance()->EnableProcessing();
for (int i = 0; i < kNumSst; i++) {
for (int j = 0; j < 100; j++) {
ASSERT_OK(Put(1, Key(i * 10 + j), "value"));
}
ASSERT_OK(Flush(1));
}
// Reopen with verification should report corruption
options.verify_sst_unique_id_in_manifest = true;
auto s = TryReopenWithColumnFamilies({"default", "one", "two"}, options);
ASSERT_TRUE(s.IsCorruption());
}
#ifndef ROCKSDB_LITE
TEST_F(DBTest2, GetLatestSeqAndTsForKey) {
Destroy(last_options_);
Options options = CurrentOptions();
options.max_write_buffer_size_to_maintain = 64 << 10;
options.create_if_missing = true;
options.disable_auto_compactions = true;
options.comparator = test::BytewiseComparatorWithU64TsWrapper();
options.statistics = CreateDBStatistics();
Reopen(options);
constexpr uint64_t kTsU64Value = 12;
for (uint64_t key = 0; key < 100; ++key) {
Revise APIs related to user-defined timestamp (#8946) Summary: ajkr reminded me that we have a rule of not including per-kv related data in `WriteOptions`. Namely, `WriteOptions` should not include information about "what-to-write", but should just include information about "how-to-write". According to this rule, `WriteOptions::timestamp` (experimental) is clearly a violation. Therefore, this PR removes `WriteOptions::timestamp` for compliance. After the removal, we need to pass timestamp info via another set of APIs. This PR proposes a set of overloaded functions `Put(write_opts, key, value, ts)`, `Delete(write_opts, key, ts)`, and `SingleDelete(write_opts, key, ts)`. Planned to add `Write(write_opts, batch, ts)`, but its complexity made me reconsider doing it in another PR (maybe). For better checking and returning error early, we also add a new set of APIs to `WriteBatch` that take extra `timestamp` information when writing to `WriteBatch`es. These set of APIs in `WriteBatchWithIndex` are currently not supported, and are on our TODO list. Removed `WriteBatch::AssignTimestamps()` and renamed `WriteBatch::AssignTimestamp()` to `WriteBatch::UpdateTimestamps()` since this method require that all keys have space for timestamps allocated already and multiple timestamps can be updated. The constructor of `WriteBatch` now takes a fourth argument `default_cf_ts_sz` which is the timestamp size of the default column family. This will be used to allocate space when calling APIs that do not specify a column family handle. Also, updated `DB::Get()`, `DB::MultiGet()`, `DB::NewIterator()`, `DB::NewIterators()` methods, replacing some assertions about timestamp to returning Status code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8946 Test Plan: make check ./db_bench -benchmarks=fillseq,fillrandom,readrandom,readseq,deleterandom -user_timestamp_size=8 ./db_stress --user_timestamp_size=8 -nooverwritepercent=0 -test_secondary=0 -secondary_catch_up_one_in=0 -continuous_verification_interval=0 Make sure there is no perf regression by running the following ``` ./db_bench_opt -db=/dev/shm/rocksdb -use_existing_db=0 -level0_stop_writes_trigger=256 -level0_slowdown_writes_trigger=256 -level0_file_num_compaction_trigger=256 -disable_wal=1 -duration=10 -benchmarks=fillrandom ``` Before this PR ``` DB path: [/dev/shm/rocksdb] fillrandom : 1.831 micros/op 546235 ops/sec; 60.4 MB/s ``` After this PR ``` DB path: [/dev/shm/rocksdb] fillrandom : 1.820 micros/op 549404 ops/sec; 60.8 MB/s ``` Reviewed By: ltamasi Differential Revision: D33721359 Pulled By: riversand963 fbshipit-source-id: c131561534272c120ffb80711d42748d21badf09
3 years ago
std::string ts;
PutFixed64(&ts, kTsU64Value);
std::string key_str;
PutFixed64(&key_str, key);
std::reverse(key_str.begin(), key_str.end());
Revise APIs related to user-defined timestamp (#8946) Summary: ajkr reminded me that we have a rule of not including per-kv related data in `WriteOptions`. Namely, `WriteOptions` should not include information about "what-to-write", but should just include information about "how-to-write". According to this rule, `WriteOptions::timestamp` (experimental) is clearly a violation. Therefore, this PR removes `WriteOptions::timestamp` for compliance. After the removal, we need to pass timestamp info via another set of APIs. This PR proposes a set of overloaded functions `Put(write_opts, key, value, ts)`, `Delete(write_opts, key, ts)`, and `SingleDelete(write_opts, key, ts)`. Planned to add `Write(write_opts, batch, ts)`, but its complexity made me reconsider doing it in another PR (maybe). For better checking and returning error early, we also add a new set of APIs to `WriteBatch` that take extra `timestamp` information when writing to `WriteBatch`es. These set of APIs in `WriteBatchWithIndex` are currently not supported, and are on our TODO list. Removed `WriteBatch::AssignTimestamps()` and renamed `WriteBatch::AssignTimestamp()` to `WriteBatch::UpdateTimestamps()` since this method require that all keys have space for timestamps allocated already and multiple timestamps can be updated. The constructor of `WriteBatch` now takes a fourth argument `default_cf_ts_sz` which is the timestamp size of the default column family. This will be used to allocate space when calling APIs that do not specify a column family handle. Also, updated `DB::Get()`, `DB::MultiGet()`, `DB::NewIterator()`, `DB::NewIterators()` methods, replacing some assertions about timestamp to returning Status code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8946 Test Plan: make check ./db_bench -benchmarks=fillseq,fillrandom,readrandom,readseq,deleterandom -user_timestamp_size=8 ./db_stress --user_timestamp_size=8 -nooverwritepercent=0 -test_secondary=0 -secondary_catch_up_one_in=0 -continuous_verification_interval=0 Make sure there is no perf regression by running the following ``` ./db_bench_opt -db=/dev/shm/rocksdb -use_existing_db=0 -level0_stop_writes_trigger=256 -level0_slowdown_writes_trigger=256 -level0_file_num_compaction_trigger=256 -disable_wal=1 -duration=10 -benchmarks=fillrandom ``` Before this PR ``` DB path: [/dev/shm/rocksdb] fillrandom : 1.831 micros/op 546235 ops/sec; 60.4 MB/s ``` After this PR ``` DB path: [/dev/shm/rocksdb] fillrandom : 1.820 micros/op 549404 ops/sec; 60.8 MB/s ``` Reviewed By: ltamasi Differential Revision: D33721359 Pulled By: riversand963 fbshipit-source-id: c131561534272c120ffb80711d42748d21badf09
3 years ago
ASSERT_OK(db_->Put(WriteOptions(), key_str, ts, "value"));
}
ASSERT_OK(Flush());
constexpr bool cache_only = true;
constexpr SequenceNumber lower_bound_seq = 0;
auto* cfhi = static_cast_with_check<ColumnFamilyHandleImpl>(
dbfull()->DefaultColumnFamily());
assert(cfhi);
assert(cfhi->cfd());
SuperVersion* sv = cfhi->cfd()->GetSuperVersion();
for (uint64_t key = 0; key < 100; ++key) {
std::string key_str;
PutFixed64(&key_str, key);
std::reverse(key_str.begin(), key_str.end());
std::string ts;
SequenceNumber seq = kMaxSequenceNumber;
bool found_record_for_key = false;
bool is_blob_index = false;
const Status s = dbfull()->GetLatestSequenceForKey(
sv, key_str, cache_only, lower_bound_seq, &seq, &ts,
&found_record_for_key, &is_blob_index);
ASSERT_OK(s);
std::string expected_ts;
PutFixed64(&expected_ts, kTsU64Value);
ASSERT_EQ(expected_ts, ts);
ASSERT_TRUE(found_record_for_key);
ASSERT_FALSE(is_blob_index);
}
// Verify that no read to SST files.
ASSERT_EQ(0, options.statistics->getTickerCount(GET_HIT_L0));
}
#endif // ROCKSDB_LITE
} // namespace ROCKSDB_NAMESPACE
int main(int argc, char** argv) {
ROCKSDB_NAMESPACE::port::InstallStackTraceHandler();
::testing::InitGoogleTest(&argc, argv);
RegisterCustomObjects(argc, argv);
return RUN_ALL_TESTS();
}