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/table_cache.cc

418 lines
15 KiB

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree. An additional grant
// of patent rights can be found in the PATENTS file in the same directory.
// This source code is also licensed under the GPLv2 license found in the
// COPYING file in the root directory of this source tree.
//
// 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 "db/table_cache.h"
#include "db/dbformat.h"
#include "db/version_edit.h"
#include "util/filename.h"
#include "monitoring/perf_context_imp.h"
#include "rocksdb/statistics.h"
#include "table/get_context.h"
#include "table/internal_iterator.h"
#include "table/iterator_wrapper.h"
#include "table/table_builder.h"
#include "table/table_reader.h"
#include "util/coding.h"
#include "util/file_reader_writer.h"
#include "util/stop_watch.h"
#include "util/sync_point.h"
namespace rocksdb {
namespace {
template <class T>
static void DeleteEntry(const Slice& key, void* value) {
T* typed_value = reinterpret_cast<T*>(value);
delete typed_value;
}
static void UnrefEntry(void* arg1, void* arg2) {
Cache* cache = reinterpret_cast<Cache*>(arg1);
Cache::Handle* h = reinterpret_cast<Cache::Handle*>(arg2);
cache->Release(h);
}
static void DeleteTableReader(void* arg1, void* arg2) {
TableReader* table_reader = reinterpret_cast<TableReader*>(arg1);
delete table_reader;
}
static Slice GetSliceForFileNumber(const uint64_t* file_number) {
return Slice(reinterpret_cast<const char*>(file_number),
sizeof(*file_number));
}
#ifndef ROCKSDB_LITE
void AppendVarint64(IterKey* key, uint64_t v) {
char buf[10];
auto ptr = EncodeVarint64(buf, v);
key->TrimAppend(key->Size(), buf, ptr - buf);
}
#endif // ROCKSDB_LITE
} // namespace
TableCache::TableCache(const ImmutableCFOptions& ioptions,
const EnvOptions& env_options, Cache* const cache)
: ioptions_(ioptions), env_options_(env_options), cache_(cache) {
if (ioptions_.row_cache) {
// If the same cache is shared by multiple instances, we need to
// disambiguate its entries.
PutVarint64(&row_cache_id_, ioptions_.row_cache->NewId());
}
}
TableCache::~TableCache() {
}
TableReader* TableCache::GetTableReaderFromHandle(Cache::Handle* handle) {
return reinterpret_cast<TableReader*>(cache_->Value(handle));
}
void TableCache::ReleaseHandle(Cache::Handle* handle) {
cache_->Release(handle);
}
Status TableCache::GetTableReader(
const EnvOptions& env_options,
const InternalKeyComparator& internal_comparator, const FileDescriptor& fd,
bool sequential_mode, size_t readahead, bool record_read_stats,
HistogramImpl* file_read_hist, unique_ptr<TableReader>* table_reader,
bool skip_filters, int level, bool prefetch_index_and_filter_in_cache) {
std::string fname =
TableFileName(ioptions_.db_paths, fd.GetNumber(), fd.GetPathId());
unique_ptr<RandomAccessFile> file;
Status s = ioptions_.env->NewRandomAccessFile(fname, &file, env_options);
RecordTick(ioptions_.statistics, NO_FILE_OPENS);
if (s.ok()) {
Fix a crash when compaction fails to open a file Summary: We've got a crash with this stack trace: Program terminated with signal SIGTRAP, Trace/breakpoint trap. #0 0x00007fc85f2f4009 in raise () from /usr/local/fbcode/gcc-4.9-glibc-2.20-fb/lib/libpthread.so.0 #1 0x00000000005c8f61 in facebook::logdevice::handle_sigsegv(int) () at logdevice/server/sigsegv.cpp:159 #2 0x00007fc85f2f4150 in <signal handler called> () at /usr/local/fbcode/gcc-4.9-glibc-2.20-fb/lib/libpthread.so.0 #3 0x00000000031ed80c in rocksdb::NewReadaheadRandomAccessFile() at util/file_reader_writer.cc:383 #4 0x00000000031ed80c in rocksdb::NewReadaheadRandomAccessFile() at util/file_reader_writer.cc:472 #5 0x00000000031558e7 in rocksdb::TableCache::GetTableReader() at db/table_cache.cc:99 #6 0x0000000003156329 in rocksdb::TableCache::NewIterator() at db/table_cache.cc:198 #7 0x0000000003166568 in rocksdb::VersionSet::MakeInputIterator() at db/version_set.cc:3345 #8 0x000000000324a94f in rocksdb::CompactionJob::ProcessKeyValueCompaction(rocksdb::CompactionJob::SubcompactionState*) () at db/compaction_job.cc:650 #9 0x000000000324c2f6 in rocksdb::CompactionJob::Run() () at db/compaction_job.cc:530 #10 0x00000000030f5ae5 in rocksdb::DBImpl::BackgroundCompaction() at db/db_impl.cc:3269 #11 0x0000000003108d36 in rocksdb::DBImpl::BackgroundCallCompaction(void*) () at db/db_impl.cc:2970 #12 0x00000000029a2a9a in facebook::logdevice::RocksDBEnv::callback(void*) () at logdevice/server/locallogstore/RocksDBEnv.cpp:26 #13 0x00000000029a2a9a in facebook::logdevice::RocksDBEnv::callback(void*) () at logdevice/server/locallogstore/RocksDBEnv.cpp:30 #14 0x00000000031e7521 in rocksdb::ThreadPool::BGThread() at util/threadpool.cc:230 #15 0x00000000031e7663 in rocksdb::BGThreadWrapper(void*) () at util/threadpool.cc:254 #16 0x00007fc85f2ea7f1 in start_thread () at /usr/local/fbcode/gcc-4.9-glibc-2.20-fb/lib/libpthread.so.0 #17 0x00007fc85e8fb46d in clone () at /usr/local/fbcode/gcc-4.9-glibc-2.20-fb/lib/libc.so.6 From looking at the code, probably what happened is this: - `TableCache::GetTableReader()` called `Env::NewRandomAccessFile()`, which dispatched to a `PosixEnv::NewRandomAccessFile()`, where probably an `open()` call failed, so the `NewRandomAccessFile()` left a nullptr in the resulting file, - `TableCache::GetTableReader()` called `NewReadaheadRandomAccessFile()` with that `nullptr` file, - it tried to call file's method and crashed. This diff is a trivial fix to this crash. Test Plan: `make -j check` Reviewers: sdong, andrewkr, IslamAbdelRahman Reviewed By: IslamAbdelRahman Subscribers: andrewkr, dhruba Differential Revision: https://reviews.facebook.net/D62451
8 years ago
if (readahead > 0) {
file = NewReadaheadRandomAccessFile(std::move(file), readahead);
}
if (!sequential_mode && ioptions_.advise_random_on_open) {
file->Hint(RandomAccessFile::RANDOM);
}
StopWatch sw(ioptions_.env, ioptions_.statistics, TABLE_OPEN_IO_MICROS);
std::unique_ptr<RandomAccessFileReader> file_reader(
new RandomAccessFileReader(std::move(file), ioptions_.env,
ioptions_.statistics, record_read_stats,
file_read_hist));
s = ioptions_.table_factory->NewTableReader(
TableReaderOptions(ioptions_, env_options, internal_comparator,
skip_filters, level),
std::move(file_reader), fd.GetFileSize(), table_reader,
prefetch_index_and_filter_in_cache);
TEST_SYNC_POINT("TableCache::GetTableReader:0");
}
return s;
}
void TableCache::EraseHandle(const FileDescriptor& fd, Cache::Handle* handle) {
ReleaseHandle(handle);
uint64_t number = fd.GetNumber();
Slice key = GetSliceForFileNumber(&number);
cache_->Erase(key);
}
Status TableCache::FindTable(const EnvOptions& env_options,
const InternalKeyComparator& internal_comparator,
const FileDescriptor& fd, Cache::Handle** handle,
const bool no_io, bool record_read_stats,
HistogramImpl* file_read_hist, bool skip_filters,
int level,
bool prefetch_index_and_filter_in_cache) {
PERF_TIMER_GUARD(find_table_nanos);
Status s;
uint64_t number = fd.GetNumber();
Slice key = GetSliceForFileNumber(&number);
*handle = cache_->Lookup(key);
TEST_SYNC_POINT_CALLBACK("TableCache::FindTable:0",
const_cast<bool*>(&no_io));
if (*handle == nullptr) {
if (no_io) { // Don't do IO and return a not-found status
return Status::Incomplete("Table not found in table_cache, no_io is set");
}
unique_ptr<TableReader> table_reader;
s = GetTableReader(env_options, internal_comparator, fd,
false /* sequential mode */, 0 /* readahead */,
record_read_stats, file_read_hist, &table_reader,
skip_filters, level, prefetch_index_and_filter_in_cache);
if (!s.ok()) {
assert(table_reader == nullptr);
RecordTick(ioptions_.statistics, NO_FILE_ERRORS);
// We do not cache error results so that if the error is transient,
// or somebody repairs the file, we recover automatically.
} else {
s = cache_->Insert(key, table_reader.get(), 1, &DeleteEntry<TableReader>,
handle);
if (s.ok()) {
// Release ownership of table reader.
table_reader.release();
}
}
}
return s;
}
InternalIterator* TableCache::NewIterator(
const ReadOptions& options, const EnvOptions& env_options,
const InternalKeyComparator& icomparator, const FileDescriptor& fd,
RangeDelAggregator* range_del_agg, TableReader** table_reader_ptr,
HistogramImpl* file_read_hist, bool for_compaction, Arena* arena,
bool skip_filters, int level) {
PERF_TIMER_GUARD(new_table_iterator_nanos);
Status s;
bool create_new_table_reader = false;
TableReader* table_reader = nullptr;
Cache::Handle* handle = nullptr;
if (s.ok()) {
if (table_reader_ptr != nullptr) {
*table_reader_ptr = nullptr;
}
size_t readahead = 0;
if (for_compaction) {
#ifndef NDEBUG
bool use_direct_reads_for_compaction = env_options.use_direct_reads;
TEST_SYNC_POINT_CALLBACK("TableCache::NewIterator:for_compaction",
&use_direct_reads_for_compaction);
#endif // !NDEBUG
if (ioptions_.new_table_reader_for_compaction_inputs) {
readahead = ioptions_.compaction_readahead_size;
create_new_table_reader = true;
}
} else {
readahead = options.readahead_size;
create_new_table_reader = readahead > 0;
}
if (create_new_table_reader) {
unique_ptr<TableReader> table_reader_unique_ptr;
s = GetTableReader(
env_options, icomparator, fd, true /* sequential_mode */, readahead,
!for_compaction /* record stats */, nullptr, &table_reader_unique_ptr,
false /* skip_filters */, level);
if (s.ok()) {
table_reader = table_reader_unique_ptr.release();
}
} else {
table_reader = fd.table_reader;
if (table_reader == nullptr) {
s = FindTable(env_options, icomparator, fd, &handle,
options.read_tier == kBlockCacheTier /* no_io */,
!for_compaction /* record read_stats */, file_read_hist,
skip_filters, level);
if (s.ok()) {
table_reader = GetTableReaderFromHandle(handle);
}
}
}
}
InternalIterator* result = nullptr;
if (s.ok()) {
result =
table_reader->NewIterator(options, arena, &icomparator, skip_filters);
if (create_new_table_reader) {
assert(handle == nullptr);
result->RegisterCleanup(&DeleteTableReader, table_reader, nullptr);
} else if (handle != nullptr) {
result->RegisterCleanup(&UnrefEntry, cache_, handle);
handle = nullptr; // prevent from releasing below
}
if (for_compaction) {
table_reader->SetupForCompaction();
}
if (table_reader_ptr != nullptr) {
*table_reader_ptr = table_reader;
}
}
if (s.ok() && range_del_agg != nullptr && !options.ignore_range_deletions) {
std::unique_ptr<InternalIterator> range_del_iter(
table_reader->NewRangeTombstoneIterator(options));
if (range_del_iter != nullptr) {
s = range_del_iter->status();
}
if (s.ok()) {
s = range_del_agg->AddTombstones(std::move(range_del_iter));
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
8 years ago
}
}
if (handle != nullptr) {
ReleaseHandle(handle);
}
if (!s.ok()) {
assert(result == nullptr);
result = NewErrorInternalIterator(s, arena);
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
8 years ago
}
return result;
}
Status TableCache::Get(const ReadOptions& options,
const InternalKeyComparator& internal_comparator,
const FileDescriptor& fd, const Slice& k,
GetContext* get_context, HistogramImpl* file_read_hist,
bool skip_filters, int level) {
std::string* row_cache_entry = nullptr;
bool done = false;
#ifndef ROCKSDB_LITE
IterKey row_cache_key;
std::string row_cache_entry_buffer;
// Check row cache if enabled. Since row cache does not currently store
// sequence numbers, we cannot use it if we need to fetch the sequence.
if (ioptions_.row_cache && !get_context->NeedToReadSequence()) {
uint64_t fd_number = fd.GetNumber();
auto user_key = ExtractUserKey(k);
// We use the user key as cache key instead of the internal key,
// otherwise the whole cache would be invalidated every time the
// sequence key increases. However, to support caching snapshot
// reads, we append the sequence number (incremented by 1 to
// distinguish from 0) only in this case.
uint64_t seq_no =
options.snapshot == nullptr ? 0 : 1 + GetInternalKeySeqno(k);
// Compute row cache key.
row_cache_key.TrimAppend(row_cache_key.Size(), row_cache_id_.data(),
row_cache_id_.size());
AppendVarint64(&row_cache_key, fd_number);
AppendVarint64(&row_cache_key, seq_no);
row_cache_key.TrimAppend(row_cache_key.Size(), user_key.data(),
user_key.size());
if (auto row_handle =
ioptions_.row_cache->Lookup(row_cache_key.GetUserKey())) {
auto found_row_cache_entry = static_cast<const std::string*>(
ioptions_.row_cache->Value(row_handle));
replayGetContextLog(*found_row_cache_entry, user_key, get_context);
ioptions_.row_cache->Release(row_handle);
RecordTick(ioptions_.statistics, ROW_CACHE_HIT);
done = true;
} else {
// Not found, setting up the replay log.
RecordTick(ioptions_.statistics, ROW_CACHE_MISS);
row_cache_entry = &row_cache_entry_buffer;
}
}
#endif // ROCKSDB_LITE
Status s;
TableReader* t = fd.table_reader;
Cache::Handle* handle = nullptr;
if (!done && s.ok()) {
if (t == nullptr) {
s = FindTable(env_options_, internal_comparator, fd, &handle,
options.read_tier == kBlockCacheTier /* no_io */,
true /* record_read_stats */, file_read_hist, skip_filters,
level);
if (s.ok()) {
t = GetTableReaderFromHandle(handle);
}
}
if (s.ok() && get_context->range_del_agg() != nullptr &&
!options.ignore_range_deletions) {
std::unique_ptr<InternalIterator> range_del_iter(
t->NewRangeTombstoneIterator(options));
if (range_del_iter != nullptr) {
s = range_del_iter->status();
}
if (s.ok()) {
s = get_context->range_del_agg()->AddTombstones(
std::move(range_del_iter));
}
}
if (s.ok()) {
get_context->SetReplayLog(row_cache_entry); // nullptr if no cache.
s = t->Get(options, k, get_context, skip_filters);
get_context->SetReplayLog(nullptr);
} else if (options.read_tier == kBlockCacheTier && s.IsIncomplete()) {
// Couldn't find Table in cache but treat as kFound if no_io set
get_context->MarkKeyMayExist();
s = Status::OK();
done = true;
}
}
#ifndef ROCKSDB_LITE
// Put the replay log in row cache only if something was found.
if (!done && s.ok() && row_cache_entry && !row_cache_entry->empty()) {
size_t charge =
row_cache_key.Size() + row_cache_entry->size() + sizeof(std::string);
void* row_ptr = new std::string(std::move(*row_cache_entry));
ioptions_.row_cache->Insert(row_cache_key.GetUserKey(), row_ptr, charge,
&DeleteEntry<std::string>);
}
#endif // ROCKSDB_LITE
if (handle != nullptr) {
ReleaseHandle(handle);
}
return s;
}
Status TableCache::GetTableProperties(
const EnvOptions& env_options,
const InternalKeyComparator& internal_comparator, const FileDescriptor& fd,
std::shared_ptr<const TableProperties>* properties, bool no_io) {
Status s;
auto table_reader = fd.table_reader;
// table already been pre-loaded?
if (table_reader) {
*properties = table_reader->GetTableProperties();
return s;
}
Cache::Handle* table_handle = nullptr;
s = FindTable(env_options, internal_comparator, fd, &table_handle, no_io);
if (!s.ok()) {
return s;
}
assert(table_handle);
auto table = GetTableReaderFromHandle(table_handle);
*properties = table->GetTableProperties();
ReleaseHandle(table_handle);
return s;
}
size_t TableCache::GetMemoryUsageByTableReader(
const EnvOptions& env_options,
const InternalKeyComparator& internal_comparator,
const FileDescriptor& fd) {
Status s;
auto table_reader = fd.table_reader;
// table already been pre-loaded?
if (table_reader) {
return table_reader->ApproximateMemoryUsage();
}
Cache::Handle* table_handle = nullptr;
s = FindTable(env_options, internal_comparator, fd, &table_handle, true);
if (!s.ok()) {
return 0;
}
assert(table_handle);
auto table = GetTableReaderFromHandle(table_handle);
auto ret = table->ApproximateMemoryUsage();
ReleaseHandle(table_handle);
return ret;
}
void TableCache::Evict(Cache* cache, uint64_t file_number) {
cache->Erase(GetSliceForFileNumber(&file_number));
}
} // namespace rocksdb