|
|
|
// 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.
|
|
|
|
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
#include "db_stress_tool/expected_state.h"
|
|
|
|
#ifdef GFLAGS
|
|
|
|
#include "db_stress_tool/db_stress_common.h"
|
|
|
|
#include "rocksdb/utilities/transaction_db.h"
|
Account memory of FileMetaData in global memory limit (#9924)
Summary:
**Context/Summary:**
As revealed by heap profiling, allocation of `FileMetaData` for [newly created file added to a Version](https://github.com/facebook/rocksdb/pull/9924/files#diff-a6aa385940793f95a2c5b39cc670bd440c4547fa54fd44622f756382d5e47e43R774) can consume significant heap memory. This PR is to account that toward our global memory limit based on block cache capacity.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9924
Test Plan:
- Previous `make check` verified there are only 2 places where the memory of the allocated `FileMetaData` can be released
- New unit test `TEST_P(ChargeFileMetadataTestWithParam, Basic)`
- db bench (CPU cost of `charge_file_metadata` in write and compact)
- **write micros/op: -0.24%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 (remove this option for pre-PR) -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 | egrep 'fillseq'`
- **compact micros/op -0.87%** : `TEST_TMPDIR=/dev/shm/testdb ./db_bench -benchmarks=fillseq -db=$TEST_TMPDIR -charge_file_metadata=1 -disable_auto_compactions=1 -write_buffer_size=100000 -num=4000000 -numdistinct=1000 && ./db_bench -benchmarks=compact -db=$TEST_TMPDIR -use_existing_db=1 -charge_file_metadata=1 -disable_auto_compactions=1 | egrep 'compact'`
table 1 - write
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 3.9711 | 0.264408 | 3.9914 | 0.254563 | 0.5111933721
20 | 3.83905 | 0.0664488 | 3.8251 | 0.0695456 | -0.3633711465
40 | 3.86625 | 0.136669 | 3.8867 | 0.143765 | 0.5289363078
80 | 3.87828 | 0.119007 | 3.86791 | 0.115674 | **-0.2673865734**
160 | 3.87677 | 0.162231 | 3.86739 | 0.16663 | **-0.2419539978**
table 2 - compact
#-run | (pre-PR) avg micros/op | std micros/op | (post-PR) micros/op | std micros/op | change (%)
-- | -- | -- | -- | -- | --
10 | 2,399,650.00 | 96,375.80 | 2,359,537.00 | 53,243.60 | -1.67
20 | 2,410,480.00 | 89,988.00 | 2,433,580.00 | 91,121.20 | 0.96
40 | 2.41E+06 | 121811 | 2.39E+06 | 131525 | **-0.96**
80 | 2.40E+06 | 134503 | 2.39E+06 | 108799 | **-0.78**
- stress test: `python3 tools/db_crashtest.py blackbox --charge_file_metadata=1 --cache_size=1` killed as normal
Reviewed By: ajkr
Differential Revision: D36055583
Pulled By: hx235
fbshipit-source-id: b60eab94707103cb1322cf815f05810ef0232625
3 years ago
|
|
|
#include "utilities/fault_injection_fs.h"
|
|
|
|
|
|
|
|
namespace ROCKSDB_NAMESPACE {
|
|
|
|
class NonBatchedOpsStressTest : public StressTest {
|
|
|
|
public:
|
|
|
|
NonBatchedOpsStressTest() {}
|
|
|
|
|
|
|
|
virtual ~NonBatchedOpsStressTest() {}
|
|
|
|
|
|
|
|
void VerifyDb(ThreadState* thread) const override {
|
|
|
|
// This `ReadOptions` is for validation purposes. Ignore
|
|
|
|
// `FLAGS_rate_limit_user_ops` to avoid slowing any validation.
|
|
|
|
ReadOptions options(FLAGS_verify_checksum, true);
|
Add user-defined timestamps to db_stress (#8061)
Summary:
Add some basic test for user-defined timestamp to db_stress. Currently,
read with timestamp always tries to read using the current timestamp.
Due to the per-key timestamp-sequence ordering constraint, we only add timestamp-
related tests to the `NonBatchedOpsStressTest` since this test serializes accesses
to the same key and uses a file to cross-check data correctness.
The timestamp feature is not supported in a number of components, e.g. Merge, SingleDelete,
DeleteRange, CompactionFilter, Readonly instance, secondary instance, SST file ingestion, transaction,
etc. Therefore, db_stress should exit if user enables both timestamp and these features at the same
time. The (currently) incompatible features can be found in
`CheckAndSetOptionsForUserTimestamp`.
This PR also fixes a bug triggered when timestamp is enabled together with
`index_type=kBinarySearchWithFirstKey`. This bug fix will also be in another separate PR
with more unit tests coverage. Fixing it here because I do not want to exclude the index type
from crash test.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/8061
Test Plan: make crash_test_with_ts
Reviewed By: jay-zhuang
Differential Revision: D27056282
Pulled By: riversand963
fbshipit-source-id: c3e00ad1023fdb9ebbdf9601ec18270c5e2925a9
4 years ago
|
|
|
std::string ts_str;
|
|
|
|
Slice ts;
|
|
|
|
if (FLAGS_user_timestamp_size > 0) {
|
|
|
|
ts_str = GetNowNanos();
|
Add user-defined timestamps to db_stress (#8061)
Summary:
Add some basic test for user-defined timestamp to db_stress. Currently,
read with timestamp always tries to read using the current timestamp.
Due to the per-key timestamp-sequence ordering constraint, we only add timestamp-
related tests to the `NonBatchedOpsStressTest` since this test serializes accesses
to the same key and uses a file to cross-check data correctness.
The timestamp feature is not supported in a number of components, e.g. Merge, SingleDelete,
DeleteRange, CompactionFilter, Readonly instance, secondary instance, SST file ingestion, transaction,
etc. Therefore, db_stress should exit if user enables both timestamp and these features at the same
time. The (currently) incompatible features can be found in
`CheckAndSetOptionsForUserTimestamp`.
This PR also fixes a bug triggered when timestamp is enabled together with
`index_type=kBinarySearchWithFirstKey`. This bug fix will also be in another separate PR
with more unit tests coverage. Fixing it here because I do not want to exclude the index type
from crash test.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/8061
Test Plan: make crash_test_with_ts
Reviewed By: jay-zhuang
Differential Revision: D27056282
Pulled By: riversand963
fbshipit-source-id: c3e00ad1023fdb9ebbdf9601ec18270c5e2925a9
4 years ago
|
|
|
ts = ts_str;
|
|
|
|
options.timestamp = &ts;
|
|
|
|
}
|
|
|
|
|
|
|
|
auto shared = thread->shared;
|
|
|
|
const int64_t max_key = shared->GetMaxKey();
|
|
|
|
const int64_t keys_per_thread = max_key / shared->GetNumThreads();
|
|
|
|
int64_t start = keys_per_thread * thread->tid;
|
|
|
|
int64_t end = start + keys_per_thread;
|
|
|
|
uint64_t prefix_to_use =
|
|
|
|
(FLAGS_prefix_size < 0) ? 1 : static_cast<size_t>(FLAGS_prefix_size);
|
|
|
|
|
|
|
|
if (thread->tid == shared->GetNumThreads() - 1) {
|
|
|
|
end = max_key;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (size_t cf = 0; cf < column_families_.size(); ++cf) {
|
|
|
|
if (thread->shared->HasVerificationFailedYet()) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
enum class VerificationMethod {
|
|
|
|
kIterator,
|
|
|
|
kGet,
|
|
|
|
kGetEntity,
|
|
|
|
kMultiGet,
|
|
|
|
kMultiGetEntity,
|
|
|
|
kGetMergeOperands,
|
|
|
|
// Add any new items above kNumberOfMethods
|
|
|
|
kNumberOfMethods
|
|
|
|
};
|
|
|
|
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
constexpr int num_methods =
|
|
|
|
static_cast<int>(VerificationMethod::kNumberOfMethods);
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
|
|
|
|
const VerificationMethod method =
|
|
|
|
static_cast<VerificationMethod>(thread->rand.Uniform(
|
|
|
|
(FLAGS_user_timestamp_size > 0) ? num_methods - 1 : num_methods));
|
|
|
|
|
|
|
|
if (method == VerificationMethod::kIterator) {
|
|
|
|
std::unique_ptr<Iterator> iter(
|
|
|
|
db_->NewIterator(options, column_families_[cf]));
|
|
|
|
|
|
|
|
std::string seek_key = Key(start);
|
|
|
|
iter->Seek(seek_key);
|
|
|
|
|
|
|
|
Slice prefix(seek_key.data(), prefix_to_use);
|
|
|
|
|
|
|
|
for (int64_t i = start; i < end; ++i) {
|
|
|
|
if (thread->shared->HasVerificationFailedYet()) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
const std::string key = Key(i);
|
|
|
|
const Slice k(key);
|
|
|
|
const Slice pfx(key.data(), prefix_to_use);
|
|
|
|
|
|
|
|
// Reseek when the prefix changes
|
|
|
|
if (prefix_to_use > 0 && prefix.compare(pfx) != 0) {
|
|
|
|
iter->Seek(k);
|
|
|
|
seek_key = key;
|
|
|
|
prefix = Slice(seek_key.data(), prefix_to_use);
|
|
|
|
}
|
|
|
|
|
|
|
|
Status s = iter->status();
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
|
|
|
|
std::string from_db;
|
|
|
|
|
|
|
|
if (iter->Valid()) {
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
const int diff = iter->key().compare(k);
|
|
|
|
|
|
|
|
if (diff > 0) {
|
|
|
|
s = Status::NotFound();
|
|
|
|
} else if (diff == 0) {
|
|
|
|
if (!VerifyWideColumns(iter->value(), iter->columns())) {
|
|
|
|
VerificationAbort(shared, static_cast<int>(cf), i,
|
|
|
|
iter->value(), iter->columns());
|
|
|
|
}
|
|
|
|
|
|
|
|
from_db = iter->value().ToString();
|
|
|
|
iter->Next();
|
|
|
|
} else {
|
|
|
|
assert(diff < 0);
|
|
|
|
|
|
|
|
VerificationAbort(shared, "An out of range key was found",
|
|
|
|
static_cast<int>(cf), i);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
// The iterator found no value for the key in question, so do not
|
|
|
|
// move to the next item in the iterator
|
|
|
|
s = Status::NotFound();
|
|
|
|
}
|
|
|
|
|
|
|
|
VerifyOrSyncValue(static_cast<int>(cf), i, options, shared, from_db,
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
/* msg_prefix */ "Iterator verification", s);
|
|
|
|
|
|
|
|
if (!from_db.empty()) {
|
|
|
|
PrintKeyValue(static_cast<int>(cf), static_cast<uint32_t>(i),
|
|
|
|
from_db.data(), from_db.size());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else if (method == VerificationMethod::kGet) {
|
|
|
|
for (int64_t i = start; i < end; ++i) {
|
|
|
|
if (thread->shared->HasVerificationFailedYet()) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
const std::string key = Key(i);
|
|
|
|
std::string from_db;
|
|
|
|
|
|
|
|
Status s = db_->Get(options, column_families_[cf], key, &from_db);
|
|
|
|
|
|
|
|
VerifyOrSyncValue(static_cast<int>(cf), i, options, shared, from_db,
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
/* msg_prefix */ "Get verification", s);
|
|
|
|
|
|
|
|
if (!from_db.empty()) {
|
|
|
|
PrintKeyValue(static_cast<int>(cf), static_cast<uint32_t>(i),
|
|
|
|
from_db.data(), from_db.size());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else if (method == VerificationMethod::kGetEntity) {
|
|
|
|
for (int64_t i = start; i < end; ++i) {
|
|
|
|
if (thread->shared->HasVerificationFailedYet()) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
const std::string key = Key(i);
|
|
|
|
PinnableWideColumns result;
|
|
|
|
|
|
|
|
Status s =
|
|
|
|
db_->GetEntity(options, column_families_[cf], key, &result);
|
|
|
|
|
|
|
|
std::string from_db;
|
|
|
|
|
|
|
|
if (s.ok()) {
|
|
|
|
const WideColumns& columns = result.columns();
|
|
|
|
|
|
|
|
if (!columns.empty() &&
|
|
|
|
columns.front().name() == kDefaultWideColumnName) {
|
|
|
|
from_db = columns.front().value().ToString();
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!VerifyWideColumns(columns)) {
|
|
|
|
VerificationAbort(shared, static_cast<int>(cf), i, from_db,
|
|
|
|
columns);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
VerifyOrSyncValue(static_cast<int>(cf), i, options, shared, from_db,
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
/* msg_prefix */ "GetEntity verification", s);
|
|
|
|
|
|
|
|
if (!from_db.empty()) {
|
|
|
|
PrintKeyValue(static_cast<int>(cf), static_cast<uint32_t>(i),
|
|
|
|
from_db.data(), from_db.size());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else if (method == VerificationMethod::kMultiGet) {
|
|
|
|
for (int64_t i = start; i < end;) {
|
|
|
|
if (thread->shared->HasVerificationFailedYet()) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Keep the batch size to some reasonable value
|
|
|
|
size_t batch_size = thread->rand.Uniform(128) + 1;
|
|
|
|
batch_size = std::min<size_t>(batch_size, end - i);
|
|
|
|
|
|
|
|
std::vector<std::string> key_strs(batch_size);
|
|
|
|
std::vector<Slice> keys(batch_size);
|
|
|
|
std::vector<PinnableSlice> values(batch_size);
|
|
|
|
std::vector<Status> statuses(batch_size);
|
|
|
|
|
|
|
|
for (size_t j = 0; j < batch_size; ++j) {
|
|
|
|
key_strs[j] = Key(i + j);
|
|
|
|
keys[j] = Slice(key_strs[j]);
|
|
|
|
}
|
|
|
|
|
|
|
|
db_->MultiGet(options, column_families_[cf], batch_size, keys.data(),
|
|
|
|
values.data(), statuses.data());
|
|
|
|
|
|
|
|
for (size_t j = 0; j < batch_size; ++j) {
|
|
|
|
const std::string from_db = values[j].ToString();
|
|
|
|
|
|
|
|
VerifyOrSyncValue(static_cast<int>(cf), i + j, options, shared,
|
|
|
|
from_db, /* msg_prefix */ "MultiGet verification",
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
statuses[j]);
|
|
|
|
|
|
|
|
if (!from_db.empty()) {
|
|
|
|
PrintKeyValue(static_cast<int>(cf), static_cast<uint32_t>(i + j),
|
|
|
|
from_db.data(), from_db.size());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
i += batch_size;
|
|
|
|
}
|
|
|
|
} else if (method == VerificationMethod::kMultiGetEntity) {
|
|
|
|
for (int64_t i = start; i < end;) {
|
|
|
|
if (thread->shared->HasVerificationFailedYet()) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Keep the batch size to some reasonable value
|
|
|
|
size_t batch_size = thread->rand.Uniform(128) + 1;
|
|
|
|
batch_size = std::min<size_t>(batch_size, end - i);
|
|
|
|
|
|
|
|
std::vector<std::string> key_strs(batch_size);
|
|
|
|
std::vector<Slice> keys(batch_size);
|
|
|
|
std::vector<PinnableWideColumns> results(batch_size);
|
|
|
|
std::vector<Status> statuses(batch_size);
|
|
|
|
|
|
|
|
for (size_t j = 0; j < batch_size; ++j) {
|
|
|
|
key_strs[j] = Key(i + j);
|
|
|
|
keys[j] = Slice(key_strs[j]);
|
|
|
|
}
|
|
|
|
|
|
|
|
db_->MultiGetEntity(options, column_families_[cf], batch_size,
|
|
|
|
keys.data(), results.data(), statuses.data());
|
|
|
|
|
|
|
|
for (size_t j = 0; j < batch_size; ++j) {
|
|
|
|
std::string from_db;
|
|
|
|
|
|
|
|
if (statuses[j].ok()) {
|
|
|
|
const WideColumns& columns = results[j].columns();
|
|
|
|
|
|
|
|
if (!columns.empty() &&
|
|
|
|
columns.front().name() == kDefaultWideColumnName) {
|
|
|
|
from_db = columns.front().value().ToString();
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!VerifyWideColumns(columns)) {
|
|
|
|
VerificationAbort(shared, static_cast<int>(cf), i, from_db,
|
|
|
|
columns);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
VerifyOrSyncValue(
|
|
|
|
static_cast<int>(cf), i + j, options, shared, from_db,
|
|
|
|
/* msg_prefix */ "MultiGetEntity verification", statuses[j]);
|
|
|
|
|
|
|
|
if (!from_db.empty()) {
|
|
|
|
PrintKeyValue(static_cast<int>(cf), static_cast<uint32_t>(i + j),
|
|
|
|
from_db.data(), from_db.size());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
i += batch_size;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
assert(method == VerificationMethod::kGetMergeOperands);
|
|
|
|
|
|
|
|
// Start off with small size that will be increased later if necessary
|
|
|
|
std::vector<PinnableSlice> values(4);
|
|
|
|
|
|
|
|
GetMergeOperandsOptions merge_operands_info;
|
|
|
|
merge_operands_info.expected_max_number_of_operands =
|
|
|
|
static_cast<int>(values.size());
|
|
|
|
|
|
|
|
for (int64_t i = start; i < end; ++i) {
|
|
|
|
if (thread->shared->HasVerificationFailedYet()) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
const std::string key = Key(i);
|
|
|
|
const Slice k(key);
|
|
|
|
std::string from_db;
|
|
|
|
int number_of_operands = 0;
|
|
|
|
|
|
|
|
Status s = db_->GetMergeOperands(options, column_families_[cf], k,
|
|
|
|
values.data(), &merge_operands_info,
|
|
|
|
&number_of_operands);
|
|
|
|
|
|
|
|
if (s.IsIncomplete()) {
|
|
|
|
// Need to resize values as there are more than values.size() merge
|
|
|
|
// operands on this key. Should only happen a few times when we
|
|
|
|
// encounter a key that had more merge operands than any key seen so
|
|
|
|
// far
|
|
|
|
values.resize(number_of_operands);
|
|
|
|
merge_operands_info.expected_max_number_of_operands =
|
|
|
|
static_cast<int>(number_of_operands);
|
|
|
|
s = db_->GetMergeOperands(options, column_families_[cf], k,
|
|
|
|
values.data(), &merge_operands_info,
|
|
|
|
&number_of_operands);
|
|
|
|
}
|
|
|
|
// Assumed here that GetMergeOperands always sets number_of_operand
|
|
|
|
if (number_of_operands) {
|
|
|
|
from_db = values[number_of_operands - 1].ToString();
|
|
|
|
}
|
|
|
|
|
|
|
|
VerifyOrSyncValue(static_cast<int>(cf), i, options, shared, from_db,
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
/* msg_prefix */ "GetMergeOperands verification",
|
|
|
|
s);
|
|
|
|
|
|
|
|
if (!from_db.empty()) {
|
|
|
|
PrintKeyValue(static_cast<int>(cf), static_cast<uint32_t>(i),
|
|
|
|
from_db.data(), from_db.size());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void ContinuouslyVerifyDb(ThreadState* thread) const override {
|
|
|
|
if (!cmp_db_) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
assert(cmp_db_);
|
|
|
|
assert(!cmp_cfhs_.empty());
|
|
|
|
Status s = cmp_db_->TryCatchUpWithPrimary();
|
|
|
|
if (!s.ok()) {
|
|
|
|
assert(false);
|
|
|
|
exit(1);
|
|
|
|
}
|
|
|
|
|
|
|
|
const auto checksum_column_family = [](Iterator* iter,
|
|
|
|
uint32_t* checksum) -> Status {
|
|
|
|
assert(nullptr != checksum);
|
|
|
|
uint32_t ret = 0;
|
|
|
|
for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
|
|
|
|
ret = crc32c::Extend(ret, iter->key().data(), iter->key().size());
|
|
|
|
ret = crc32c::Extend(ret, iter->value().data(), iter->value().size());
|
|
|
|
}
|
|
|
|
*checksum = ret;
|
|
|
|
return iter->status();
|
|
|
|
};
|
|
|
|
|
|
|
|
auto* shared = thread->shared;
|
|
|
|
assert(shared);
|
|
|
|
const int64_t max_key = shared->GetMaxKey();
|
|
|
|
ReadOptions read_opts(FLAGS_verify_checksum, true);
|
|
|
|
std::string ts_str;
|
|
|
|
Slice ts;
|
|
|
|
if (FLAGS_user_timestamp_size > 0) {
|
|
|
|
ts_str = GetNowNanos();
|
|
|
|
ts = ts_str;
|
|
|
|
read_opts.timestamp = &ts;
|
|
|
|
}
|
|
|
|
|
|
|
|
static Random64 rand64(shared->GetSeed());
|
|
|
|
|
|
|
|
{
|
|
|
|
uint32_t crc = 0;
|
|
|
|
std::unique_ptr<Iterator> it(cmp_db_->NewIterator(read_opts));
|
|
|
|
s = checksum_column_family(it.get(), &crc);
|
|
|
|
if (!s.ok()) {
|
|
|
|
fprintf(stderr, "Computing checksum of default cf: %s\n",
|
|
|
|
s.ToString().c_str());
|
|
|
|
assert(false);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
for (auto* handle : cmp_cfhs_) {
|
|
|
|
if (thread->rand.OneInOpt(3)) {
|
|
|
|
// Use Get()
|
|
|
|
uint64_t key = rand64.Uniform(static_cast<uint64_t>(max_key));
|
|
|
|
std::string key_str = Key(key);
|
|
|
|
std::string value;
|
|
|
|
std::string key_ts;
|
|
|
|
s = cmp_db_->Get(read_opts, handle, key_str, &value,
|
|
|
|
FLAGS_user_timestamp_size > 0 ? &key_ts : nullptr);
|
|
|
|
s.PermitUncheckedError();
|
|
|
|
} else {
|
|
|
|
// Use range scan
|
|
|
|
std::unique_ptr<Iterator> iter(cmp_db_->NewIterator(read_opts, handle));
|
|
|
|
uint32_t rnd = (thread->rand.Next()) % 4;
|
|
|
|
if (0 == rnd) {
|
|
|
|
// SeekToFirst() + Next()*5
|
|
|
|
read_opts.total_order_seek = true;
|
|
|
|
iter->SeekToFirst();
|
|
|
|
for (int i = 0; i < 5 && iter->Valid(); ++i, iter->Next()) {
|
|
|
|
}
|
|
|
|
} else if (1 == rnd) {
|
|
|
|
// SeekToLast() + Prev()*5
|
|
|
|
read_opts.total_order_seek = true;
|
|
|
|
iter->SeekToLast();
|
|
|
|
for (int i = 0; i < 5 && iter->Valid(); ++i, iter->Prev()) {
|
|
|
|
}
|
|
|
|
} else if (2 == rnd) {
|
|
|
|
// Seek() +Next()*5
|
|
|
|
uint64_t key = rand64.Uniform(static_cast<uint64_t>(max_key));
|
|
|
|
std::string key_str = Key(key);
|
|
|
|
iter->Seek(key_str);
|
|
|
|
for (int i = 0; i < 5 && iter->Valid(); ++i, iter->Next()) {
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
// SeekForPrev() + Prev()*5
|
|
|
|
uint64_t key = rand64.Uniform(static_cast<uint64_t>(max_key));
|
|
|
|
std::string key_str = Key(key);
|
|
|
|
iter->SeekForPrev(key_str);
|
|
|
|
for (int i = 0; i < 5 && iter->Valid(); ++i, iter->Prev()) {
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void MaybeClearOneColumnFamily(ThreadState* thread) override {
|
|
|
|
if (FLAGS_column_families > 1) {
|
|
|
|
if (thread->rand.OneInOpt(FLAGS_clear_column_family_one_in)) {
|
|
|
|
// drop column family and then create it again (can't drop default)
|
|
|
|
int cf = thread->rand.Next() % (FLAGS_column_families - 1) + 1;
|
|
|
|
std::string new_name =
|
|
|
|
std::to_string(new_column_family_name_.fetch_add(1));
|
|
|
|
{
|
|
|
|
MutexLock l(thread->shared->GetMutex());
|
|
|
|
fprintf(
|
|
|
|
stdout,
|
|
|
|
"[CF %d] Dropping and recreating column family. new name: %s\n",
|
|
|
|
cf, new_name.c_str());
|
|
|
|
}
|
|
|
|
thread->shared->LockColumnFamily(cf);
|
|
|
|
Status s = db_->DropColumnFamily(column_families_[cf]);
|
|
|
|
delete column_families_[cf];
|
|
|
|
if (!s.ok()) {
|
|
|
|
fprintf(stderr, "dropping column family error: %s\n",
|
|
|
|
s.ToString().c_str());
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
s = db_->CreateColumnFamily(ColumnFamilyOptions(options_), new_name,
|
|
|
|
&column_families_[cf]);
|
|
|
|
column_family_names_[cf] = new_name;
|
|
|
|
thread->shared->ClearColumnFamily(cf);
|
|
|
|
if (!s.ok()) {
|
|
|
|
fprintf(stderr, "creating column family error: %s\n",
|
|
|
|
s.ToString().c_str());
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
thread->shared->UnlockColumnFamily(cf);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
bool ShouldAcquireMutexOnKey() const override { return true; }
|
|
|
|
|
|
|
|
bool IsStateTracked() const override { return true; }
|
|
|
|
|
|
|
|
Status TestGet(ThreadState* thread, const ReadOptions& read_opts,
|
|
|
|
const std::vector<int>& rand_column_families,
|
|
|
|
const std::vector<int64_t>& rand_keys) override {
|
|
|
|
auto cfh = column_families_[rand_column_families[0]];
|
|
|
|
std::string key_str = Key(rand_keys[0]);
|
|
|
|
Slice key = key_str;
|
|
|
|
std::string from_db;
|
|
|
|
int error_count = 0;
|
|
|
|
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
fault_fs_guard->EnableErrorInjection();
|
|
|
|
SharedState::ignore_read_error = false;
|
|
|
|
}
|
|
|
|
|
|
|
|
ReadOptions read_opts_copy = read_opts;
|
|
|
|
std::string read_ts_str;
|
|
|
|
Slice read_ts_slice;
|
|
|
|
if (FLAGS_user_timestamp_size > 0) {
|
|
|
|
read_ts_str = GetNowNanos();
|
|
|
|
read_ts_slice = read_ts_str;
|
|
|
|
read_opts_copy.timestamp = &read_ts_slice;
|
|
|
|
}
|
|
|
|
bool read_older_ts = MaybeUseOlderTimestampForPointLookup(
|
|
|
|
thread, read_ts_str, read_ts_slice, read_opts_copy);
|
|
|
|
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
const ExpectedValue pre_read_expected_value =
|
|
|
|
thread->shared->Get(rand_column_families[0], rand_keys[0]);
|
|
|
|
Status s = db_->Get(read_opts_copy, cfh, key, &from_db);
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
const ExpectedValue post_read_expected_value =
|
|
|
|
thread->shared->Get(rand_column_families[0], rand_keys[0]);
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
error_count = fault_fs_guard->GetAndResetErrorCount();
|
|
|
|
}
|
|
|
|
if (s.ok()) {
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
if (error_count && !SharedState::ignore_read_error) {
|
|
|
|
// Grab mutex so multiple thread don't try to print the
|
|
|
|
// stack trace at the same time
|
|
|
|
MutexLock l(thread->shared->GetMutex());
|
|
|
|
fprintf(stderr, "Didn't get expected error from Get\n");
|
|
|
|
fprintf(stderr, "Callstack that injected the fault\n");
|
|
|
|
fault_fs_guard->PrintFaultBacktrace();
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
// found case
|
|
|
|
thread->stats.AddGets(1, 1);
|
|
|
|
// we only have the latest expected state
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
if (!FLAGS_skip_verifydb && !read_older_ts) {
|
|
|
|
if (ExpectedValueHelper::MustHaveNotExisted(pre_read_expected_value,
|
|
|
|
post_read_expected_value)) {
|
|
|
|
thread->shared->SetVerificationFailure();
|
|
|
|
fprintf(stderr,
|
|
|
|
"error : inconsistent values for key %s: Get returns %s, "
|
|
|
|
"but expected state is \"deleted\".\n",
|
|
|
|
key.ToString(true).c_str(), StringToHex(from_db).c_str());
|
|
|
|
}
|
|
|
|
Slice from_db_slice(from_db);
|
|
|
|
uint32_t value_base_from_db = GetValueBase(from_db_slice);
|
|
|
|
if (!ExpectedValueHelper::InExpectedValueBaseRange(
|
|
|
|
value_base_from_db, pre_read_expected_value,
|
|
|
|
post_read_expected_value)) {
|
|
|
|
thread->shared->SetVerificationFailure();
|
|
|
|
fprintf(stderr,
|
|
|
|
"error : inconsistent values for key %s: Get returns %s with "
|
|
|
|
"value base %d that falls out of expected state's value base "
|
|
|
|
"range.\n",
|
|
|
|
key.ToString(true).c_str(), StringToHex(from_db).c_str(),
|
|
|
|
value_base_from_db);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else if (s.IsNotFound()) {
|
|
|
|
// not found case
|
|
|
|
thread->stats.AddGets(1, 0);
|
|
|
|
if (!FLAGS_skip_verifydb && !read_older_ts) {
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
if (ExpectedValueHelper::MustHaveExisted(pre_read_expected_value,
|
|
|
|
post_read_expected_value)) {
|
|
|
|
thread->shared->SetVerificationFailure();
|
|
|
|
fprintf(stderr,
|
|
|
|
"error : inconsistent values for key %s: expected state has "
|
|
|
|
"the key, Get() returns NotFound.\n",
|
|
|
|
key.ToString(true).c_str());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if (error_count == 0) {
|
|
|
|
// errors case
|
|
|
|
thread->stats.AddErrors(1);
|
|
|
|
} else {
|
|
|
|
thread->stats.AddVerifiedErrors(1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
fault_fs_guard->DisableErrorInjection();
|
|
|
|
}
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::vector<Status> TestMultiGet(
|
|
|
|
ThreadState* thread, const ReadOptions& read_opts,
|
|
|
|
const std::vector<int>& rand_column_families,
|
|
|
|
const std::vector<int64_t>& rand_keys) override {
|
|
|
|
size_t num_keys = rand_keys.size();
|
|
|
|
std::vector<std::string> key_str;
|
|
|
|
std::vector<Slice> keys;
|
|
|
|
key_str.reserve(num_keys);
|
|
|
|
keys.reserve(num_keys);
|
|
|
|
std::vector<PinnableSlice> values(num_keys);
|
|
|
|
std::vector<Status> statuses(num_keys);
|
|
|
|
// When Flags_use_txn is enabled, we also do a read your write check.
|
|
|
|
std::vector<std::optional<ExpectedValue>> ryw_expected_values;
|
|
|
|
ryw_expected_values.reserve(num_keys);
|
|
|
|
|
|
|
|
SharedState* shared = thread->shared;
|
|
|
|
|
|
|
|
int column_family = rand_column_families[0];
|
|
|
|
ColumnFamilyHandle* cfh = column_families_[column_family];
|
|
|
|
int error_count = 0;
|
|
|
|
// Do a consistency check between Get and MultiGet. Don't do it too
|
|
|
|
// often as it will slow db_stress down
|
|
|
|
bool do_consistency_check = thread->rand.OneIn(4);
|
|
|
|
|
|
|
|
ReadOptions readoptionscopy = read_opts;
|
|
|
|
if (do_consistency_check) {
|
|
|
|
readoptionscopy.snapshot = db_->GetSnapshot();
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string read_ts_str;
|
|
|
|
Slice read_ts_slice;
|
|
|
|
MaybeUseOlderTimestampForPointLookup(thread, read_ts_str, read_ts_slice,
|
|
|
|
readoptionscopy);
|
|
|
|
|
|
|
|
readoptionscopy.rate_limiter_priority =
|
|
|
|
FLAGS_rate_limit_user_ops ? Env::IO_USER : Env::IO_TOTAL;
|
|
|
|
|
|
|
|
// To appease clang analyzer
|
|
|
|
const bool use_txn = FLAGS_use_txn;
|
|
|
|
|
|
|
|
// Create a transaction in order to write some data. The purpose is to
|
|
|
|
// exercise WriteBatchWithIndex::MultiGetFromBatchAndDB. The transaction
|
|
|
|
// will be rolled back once MultiGet returns.
|
|
|
|
Transaction* txn = nullptr;
|
|
|
|
if (use_txn) {
|
|
|
|
WriteOptions wo;
|
Rate-limit automatic WAL flush after each user write (#9607)
Summary:
**Context:**
WAL flush is currently not rate-limited by `Options::rate_limiter`. This PR is to provide rate-limiting to auto WAL flush, the one that automatically happen after each user write operation (i.e, `Options::manual_wal_flush == false`), by adding `WriteOptions::rate_limiter_options`.
Note that we are NOT rate-limiting WAL flush that do NOT automatically happen after each user write, such as `Options::manual_wal_flush == true + manual FlushWAL()` (rate-limiting multiple WAL flushes), for the benefits of:
- being consistent with [ReadOptions::rate_limiter_priority](https://github.com/facebook/rocksdb/blob/7.0.fb/include/rocksdb/options.h#L515)
- being able to turn off some WAL flush's rate-limiting but not all (e.g, turn off specific the WAL flush of a critical user write like a service's heartbeat)
`WriteOptions::rate_limiter_options` only accept `Env::IO_USER` and `Env::IO_TOTAL` currently due to an implementation constraint.
- The constraint is that we currently queue parallel writes (including WAL writes) based on FIFO policy which does not factor rate limiter priority into this layer's scheduling. If we allow lower priorities such as `Env::IO_HIGH/MID/LOW` and such writes specified with lower priorities occurs before ones specified with higher priorities (even just by a tiny bit in arrival time), the former would have blocked the latter, leading to a "priority inversion" issue and contradictory to what we promise for rate-limiting priority. Therefore we only allow `Env::IO_USER` and `Env::IO_TOTAL` right now before improving that scheduling.
A pre-requisite to this feature is to support operation-level rate limiting in `WritableFileWriter`, which is also included in this PR.
**Summary:**
- Renamed test suite `DBRateLimiterTest to DBRateLimiterOnReadTest` for adding a new test suite
- Accept `rate_limiter_priority` in `WritableFileWriter`'s private and public write functions
- Passed `WriteOptions::rate_limiter_options` to `WritableFileWriter` in the path of automatic WAL flush.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9607
Test Plan:
- Added new unit test to verify existing flush/compaction rate-limiting does not break, since `DBTest, RateLimitingTest` is disabled and current db-level rate-limiting tests focus on read only (e.g, `db_rate_limiter_test`, `DBTest2, RateLimitedCompactionReads`).
- Added new unit test `DBRateLimiterOnWriteWALTest, AutoWalFlush`
- `strace -ftt -e trace=write ./db_bench -benchmarks=fillseq -db=/dev/shm/testdb -rate_limit_auto_wal_flush=1 -rate_limiter_bytes_per_sec=15 -rate_limiter_refill_period_us=1000000 -write_buffer_size=100000000 -disable_auto_compactions=1 -num=100`
- verified that WAL flush(i.e, system-call _write_) were chunked into 15 bytes and each _write_ was roughly 1 second apart
- verified the chunking disappeared when `-rate_limit_auto_wal_flush=0`
- crash test: `python3 tools/db_crashtest.py blackbox --disable_wal=0 --rate_limit_auto_wal_flush=1 --rate_limiter_bytes_per_sec=10485760 --interval=10` killed as normal
**Benchmarked on flush/compaction to ensure no performance regression:**
- compaction with rate-limiting (see table 1, avg over 1280-run): pre-change: **915635 micros/op**; post-change:
**907350 micros/op (improved by 0.106%)**
```
#!/bin/bash
TEST_TMPDIR=/dev/shm/testdb
START=1
NUM_DATA_ENTRY=8
N=10
rm -f compact_bmk_output.txt compact_bmk_output_2.txt dont_care_output.txt
for i in $(eval echo "{$START..$NUM_DATA_ENTRY}")
do
NUM_RUN=$(($N*(2**($i-1))))
for j in $(eval echo "{$START..$NUM_RUN}")
do
./db_bench --benchmarks=fillrandom -db=$TEST_TMPDIR -disable_auto_compactions=1 -write_buffer_size=6710886 > dont_care_output.txt && ./db_bench --benchmarks=compact -use_existing_db=1 -db=$TEST_TMPDIR -level0_file_num_compaction_trigger=1 -rate_limiter_bytes_per_sec=100000000 | egrep 'compact'
done > compact_bmk_output.txt && awk -v NUM_RUN=$NUM_RUN '{sum+=$3;sum_sqrt+=$3^2}END{print sum/NUM_RUN, sqrt(sum_sqrt/NUM_RUN-(sum/NUM_RUN)^2)}' compact_bmk_output.txt >> compact_bmk_output_2.txt
done
```
- compaction w/o rate-limiting (see table 2, avg over 640-run): pre-change: **822197 micros/op**; post-change: **823148 micros/op (regressed by 0.12%)**
```
Same as above script, except that -rate_limiter_bytes_per_sec=0
```
- flush with rate-limiting (see table 3, avg over 320-run, run on the [patch](https://github.com/hx235/rocksdb/commit/ee5c6023a9f6533fab9afdc681568daa21da4953) to augment current db_bench ): pre-change: **745752 micros/op**; post-change: **745331 micros/op (regressed by 0.06 %)**
```
#!/bin/bash
TEST_TMPDIR=/dev/shm/testdb
START=1
NUM_DATA_ENTRY=8
N=10
rm -f flush_bmk_output.txt flush_bmk_output_2.txt
for i in $(eval echo "{$START..$NUM_DATA_ENTRY}")
do
NUM_RUN=$(($N*(2**($i-1))))
for j in $(eval echo "{$START..$NUM_RUN}")
do
./db_bench -db=$TEST_TMPDIR -write_buffer_size=1048576000 -num=1000000 -rate_limiter_bytes_per_sec=100000000 -benchmarks=fillseq,flush | egrep 'flush'
done > flush_bmk_output.txt && awk -v NUM_RUN=$NUM_RUN '{sum+=$3;sum_sqrt+=$3^2}END{print sum/NUM_RUN, sqrt(sum_sqrt/NUM_RUN-(sum/NUM_RUN)^2)}' flush_bmk_output.txt >> flush_bmk_output_2.txt
done
```
- flush w/o rate-limiting (see table 4, avg over 320-run, run on the [patch](https://github.com/hx235/rocksdb/commit/ee5c6023a9f6533fab9afdc681568daa21da4953) to augment current db_bench): pre-change: **487512 micros/op**, post-change: **485856 micors/ops (improved by 0.34%)**
```
Same as above script, except that -rate_limiter_bytes_per_sec=0
```
| table 1 - compact with rate-limiting|
#-run | (pre-change) avg micros/op | std micros/op | (post-change) avg micros/op | std micros/op | change in avg micros/op (%)
-- | -- | -- | -- | -- | --
10 | 896978 | 16046.9 | 901242 | 15670.9 | 0.475373978
20 | 893718 | 15813 | 886505 | 17544.7 | -0.8070778478
40 | 900426 | 23882.2 | 894958 | 15104.5 | -0.6072681153
80 | 906635 | 21761.5 | 903332 | 23948.3 | -0.3643141948
160 | 898632 | 21098.9 | 907583 | 21145 | 0.9960695813
3.20E+02 | 905252 | 22785.5 | 908106 | 25325.5 | 0.3152713278
6.40E+02 | 905213 | 23598.6 | 906741 | 21370.5 | 0.1688000504
**1.28E+03** | **908316** | **23533.1** | **907350** | **24626.8** | **-0.1063506533**
average over #-run | 901896.25 | 21064.9625 | 901977.125 | 20592.025 | 0.008967217682
| table 2 - compact w/o rate-limiting|
#-run | (pre-change) avg micros/op | std micros/op | (post-change) avg micros/op | std micros/op | change in avg micros/op (%)
-- | -- | -- | -- | -- | --
10 | 811211 | 26996.7 | 807586 | 28456.4 | -0.4468627768
20 | 815465 | 14803.7 | 814608 | 28719.7 | -0.105093413
40 | 809203 | 26187.1 | 797835 | 25492.1 | -1.404839082
80 | 822088 | 28765.3 | 822192 | 32840.4 | 0.01265071379
160 | 821719 | 36344.7 | 821664 | 29544.9 | -0.006693285661
3.20E+02 | 820921 | 27756.4 | 821403 | 28347.7 | 0.05871454135
**6.40E+02** | **822197** | **28960.6** | **823148** | **30055.1** | **0.1156657103**
average over #-run | 8.18E+05 | 2.71E+04 | 8.15E+05 | 2.91E+04 | -0.25
| table 3 - flush with rate-limiting|
#-run | (pre-change) avg micros/op | std micros/op | (post-change) avg micros/op | std micros/op | change in avg micros/op (%)
-- | -- | -- | -- | -- | --
10 | 741721 | 11770.8 | 740345 | 5949.76 | -0.1855144994
20 | 735169 | 3561.83 | 743199 | 9755.77 | 1.09226586
40 | 743368 | 8891.03 | 742102 | 8683.22 | -0.1703059588
80 | 742129 | 8148.51 | 743417 | 9631.58| 0.1735547324
160 | 749045 | 9757.21 | 746256 | 9191.86 | -0.3723407806
**3.20E+02** | **745752** | **9819.65** | **745331** | **9840.62** | **-0.0564530836**
6.40E+02 | 749006 | 11080.5 | 748173 | 10578.7 | -0.1112140624
average over #-run | 743741.4286 | 9004.218571 | 744117.5714 | 9090.215714 | 0.05057441238
| table 4 - flush w/o rate-limiting|
#-run | (pre-change) avg micros/op | std micros/op | (post-change) avg micros/op | std micros/op | change in avg micros/op (%)
-- | -- | -- | -- | -- | --
10 | 477283 | 24719.6 | 473864 | 12379 | -0.7163464863
20 | 486743 | 20175.2 | 502296 | 23931.3 | 3.195320734
40 | 482846 | 15309.2 | 489820 | 22259.5 | 1.444352858
80 | 491490 | 21883.1 | 490071 | 23085.7 | -0.2887139108
160 | 493347 | 28074.3 | 483609 | 21211.7 | -1.973864238
**3.20E+02** | **487512** | **21401.5** | **485856** | **22195.2** | **-0.3396839462**
6.40E+02 | 490307 | 25418.6 | 485435 | 22405.2 | -0.9936631539
average over #-run | 4.87E+05 | 2.24E+04 | 4.87E+05 | 2.11E+04 | 0.00E+00
Reviewed By: ajkr
Differential Revision: D34442441
Pulled By: hx235
fbshipit-source-id: 4790f13e1e5c0a95ae1d1cc93ffcf69dc6e78bdd
3 years ago
|
|
|
if (FLAGS_rate_limit_auto_wal_flush) {
|
|
|
|
wo.rate_limiter_priority = Env::IO_USER;
|
|
|
|
}
|
|
|
|
Status s = NewTxn(wo, &txn);
|
|
|
|
if (!s.ok()) {
|
|
|
|
fprintf(stderr, "NewTxn: %s\n", s.ToString().c_str());
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
for (size_t i = 0; i < num_keys; ++i) {
|
|
|
|
uint64_t rand_key = rand_keys[i];
|
|
|
|
key_str.emplace_back(Key(rand_key));
|
|
|
|
keys.emplace_back(key_str.back());
|
|
|
|
if (use_txn) {
|
|
|
|
if (!shared->AllowsOverwrite(rand_key) &&
|
|
|
|
shared->Exists(column_family, rand_key)) {
|
|
|
|
// Just do read your write checks for keys that allow overwrites.
|
|
|
|
ryw_expected_values.push_back(std::nullopt);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
// With a 1 in 10 probability, insert the just added key in the batch
|
|
|
|
// into the transaction. This will create an overlap with the MultiGet
|
|
|
|
// keys and exercise some corner cases in the code
|
|
|
|
if (thread->rand.OneIn(10)) {
|
|
|
|
int op = thread->rand.Uniform(2);
|
|
|
|
Status s;
|
|
|
|
assert(txn);
|
|
|
|
switch (op) {
|
|
|
|
case 0:
|
|
|
|
case 1: {
|
|
|
|
ExpectedValue put_value;
|
|
|
|
put_value.Put(false /* pending */);
|
|
|
|
ryw_expected_values.emplace_back(put_value);
|
|
|
|
char value[100];
|
|
|
|
size_t sz =
|
|
|
|
GenerateValue(put_value.GetValueBase(), value, sizeof(value));
|
|
|
|
Slice v(value, sz);
|
|
|
|
if (op == 0) {
|
|
|
|
s = txn->Put(cfh, keys.back(), v);
|
|
|
|
} else {
|
|
|
|
s = txn->Merge(cfh, keys.back(), v);
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case 2: {
|
|
|
|
ExpectedValue delete_value;
|
|
|
|
delete_value.Delete(false /* pending */);
|
|
|
|
ryw_expected_values.emplace_back(delete_value);
|
|
|
|
s = txn->Delete(cfh, keys.back());
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
assert(false);
|
|
|
|
}
|
|
|
|
if (!s.ok()) {
|
|
|
|
fprintf(stderr, "Transaction put: %s\n", s.ToString().c_str());
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
ryw_expected_values.push_back(std::nullopt);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!use_txn) {
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
fault_fs_guard->EnableErrorInjection();
|
|
|
|
SharedState::ignore_read_error = false;
|
|
|
|
}
|
|
|
|
db_->MultiGet(readoptionscopy, cfh, num_keys, keys.data(), values.data(),
|
|
|
|
statuses.data());
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
error_count = fault_fs_guard->GetAndResetErrorCount();
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
assert(txn);
|
|
|
|
txn->MultiGet(readoptionscopy, cfh, num_keys, keys.data(), values.data(),
|
|
|
|
statuses.data());
|
|
|
|
}
|
|
|
|
|
|
|
|
if (fault_fs_guard && error_count && !SharedState::ignore_read_error) {
|
|
|
|
int stat_nok = 0;
|
|
|
|
for (const auto& s : statuses) {
|
|
|
|
if (!s.ok() && !s.IsNotFound()) {
|
|
|
|
stat_nok++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (stat_nok < error_count) {
|
|
|
|
// Grab mutex so multiple thread don't try to print the
|
|
|
|
// stack trace at the same time
|
|
|
|
MutexLock l(thread->shared->GetMutex());
|
|
|
|
fprintf(stderr, "Didn't get expected error from MultiGet. \n");
|
|
|
|
fprintf(stderr, "num_keys %zu Expected %d errors, seen %d\n", num_keys,
|
|
|
|
error_count, stat_nok);
|
|
|
|
fprintf(stderr, "Callstack that injected the fault\n");
|
|
|
|
fault_fs_guard->PrintFaultBacktrace();
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
fault_fs_guard->DisableErrorInjection();
|
|
|
|
}
|
|
|
|
|
|
|
|
auto ryw_check =
|
|
|
|
[](const Slice& key, const PinnableSlice& value, const Status& s,
|
|
|
|
const std::optional<ExpectedValue>& ryw_expected_value) -> bool {
|
|
|
|
if (!ryw_expected_value.has_value()) {
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
const ExpectedValue& expected = ryw_expected_value.value();
|
|
|
|
char expected_value[100];
|
|
|
|
if (s.ok() &&
|
|
|
|
ExpectedValueHelper::MustHaveNotExisted(expected, expected)) {
|
|
|
|
fprintf(stderr,
|
|
|
|
"MultiGet returned value different from what was "
|
|
|
|
"written for key %s\n",
|
|
|
|
key.ToString(true).c_str());
|
|
|
|
fprintf(stderr,
|
|
|
|
"MultiGet returned ok, transaction has non-committed "
|
|
|
|
"delete.\n");
|
|
|
|
return false;
|
|
|
|
} else if (s.IsNotFound() &&
|
|
|
|
ExpectedValueHelper::MustHaveExisted(expected, expected)) {
|
|
|
|
fprintf(stderr,
|
|
|
|
"MultiGet returned value different from what was "
|
|
|
|
"written for key %s\n",
|
|
|
|
key.ToString(true).c_str());
|
|
|
|
fprintf(stderr,
|
|
|
|
"MultiGet returned not found, transaction has "
|
|
|
|
"non-committed value.\n");
|
|
|
|
return false;
|
|
|
|
} else if (s.ok() &&
|
|
|
|
ExpectedValueHelper::MustHaveExisted(expected, expected)) {
|
|
|
|
Slice from_txn_slice(value);
|
|
|
|
size_t sz = GenerateValue(expected.GetValueBase(), expected_value,
|
|
|
|
sizeof(expected_value));
|
|
|
|
Slice expected_value_slice(expected_value, sz);
|
|
|
|
if (expected_value_slice.compare(from_txn_slice) == 0) {
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
fprintf(stderr,
|
|
|
|
"MultiGet returned value different from what was "
|
|
|
|
"written for key %s\n",
|
|
|
|
key.ToString(true /* hex */).c_str());
|
|
|
|
fprintf(stderr, "MultiGet returned value %s\n",
|
|
|
|
from_txn_slice.ToString(true /* hex */).c_str());
|
|
|
|
fprintf(stderr, "Transaction has non-committed value %s\n",
|
|
|
|
expected_value_slice.ToString(true /* hex */).c_str());
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
return true;
|
|
|
|
};
|
|
|
|
|
|
|
|
auto check_multiget =
|
|
|
|
[&](const Slice& key, const PinnableSlice& expected_value,
|
|
|
|
const Status& s,
|
|
|
|
const std::optional<ExpectedValue>& ryw_expected_value) -> bool {
|
|
|
|
bool is_consistent = true;
|
|
|
|
bool is_ryw_correct = true;
|
|
|
|
// Only do the consistency check if no error was injected and
|
|
|
|
// MultiGet didn't return an unexpected error. If test does not use
|
|
|
|
// transaction, the consistency check for each key included check results
|
|
|
|
// from db `Get` and db `MultiGet` are consistent.
|
|
|
|
// If test use transaction, after consistency check, also do a read your
|
|
|
|
// own write check.
|
|
|
|
if (do_consistency_check && !error_count && (s.ok() || s.IsNotFound())) {
|
|
|
|
Status tmp_s;
|
|
|
|
std::string value;
|
|
|
|
|
|
|
|
if (use_txn) {
|
|
|
|
assert(txn);
|
|
|
|
tmp_s = txn->Get(readoptionscopy, cfh, key, &value);
|
|
|
|
} else {
|
|
|
|
tmp_s = db_->Get(readoptionscopy, cfh, key, &value);
|
|
|
|
}
|
|
|
|
if (!tmp_s.ok() && !tmp_s.IsNotFound()) {
|
|
|
|
fprintf(stderr, "Get error: %s\n", s.ToString().c_str());
|
|
|
|
is_consistent = false;
|
|
|
|
} else if (!s.ok() && tmp_s.ok()) {
|
|
|
|
fprintf(stderr, "MultiGet returned different results with key %s\n",
|
|
|
|
key.ToString(true).c_str());
|
|
|
|
fprintf(stderr, "Get returned ok, MultiGet returned not found\n");
|
|
|
|
is_consistent = false;
|
|
|
|
} else if (s.ok() && tmp_s.IsNotFound()) {
|
|
|
|
fprintf(stderr, "MultiGet returned different results with key %s\n",
|
|
|
|
key.ToString(true).c_str());
|
|
|
|
fprintf(stderr, "MultiGet returned ok, Get returned not found\n");
|
|
|
|
is_consistent = false;
|
|
|
|
} else if (s.ok() && value != expected_value.ToString()) {
|
|
|
|
fprintf(stderr, "MultiGet returned different results with key %s\n",
|
|
|
|
key.ToString(true).c_str());
|
|
|
|
fprintf(stderr, "MultiGet returned value %s\n",
|
|
|
|
expected_value.ToString(true).c_str());
|
|
|
|
fprintf(stderr, "Get returned value %s\n",
|
|
|
|
Slice(value).ToString(true /* hex */).c_str());
|
|
|
|
is_consistent = false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// If test uses transaction, continue to do a read your own write check.
|
|
|
|
if (is_consistent && use_txn) {
|
|
|
|
is_ryw_correct = ryw_check(key, expected_value, s, ryw_expected_value);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!is_consistent) {
|
|
|
|
fprintf(stderr, "TestMultiGet error: is_consistent is false\n");
|
|
|
|
thread->stats.AddErrors(1);
|
|
|
|
// Fail fast to preserve the DB state
|
|
|
|
thread->shared->SetVerificationFailure();
|
|
|
|
return false;
|
|
|
|
} else if (!is_ryw_correct) {
|
|
|
|
fprintf(stderr, "TestMultiGet error: is_ryw_correct is false\n");
|
|
|
|
thread->stats.AddErrors(1);
|
|
|
|
// Fail fast to preserve the DB state
|
|
|
|
thread->shared->SetVerificationFailure();
|
|
|
|
return false;
|
|
|
|
} else if (s.ok()) {
|
|
|
|
// found case
|
|
|
|
thread->stats.AddGets(1, 1);
|
|
|
|
} else if (s.IsNotFound()) {
|
|
|
|
// not found case
|
|
|
|
thread->stats.AddGets(1, 0);
|
|
|
|
} else if (s.IsMergeInProgress() && use_txn) {
|
|
|
|
// With txn this is sometimes expected.
|
|
|
|
thread->stats.AddGets(1, 1);
|
|
|
|
} else {
|
|
|
|
if (error_count == 0) {
|
|
|
|
// errors case
|
|
|
|
fprintf(stderr, "MultiGet error: %s\n", s.ToString().c_str());
|
|
|
|
thread->stats.AddErrors(1);
|
|
|
|
} else {
|
|
|
|
thread->stats.AddVerifiedErrors(1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return true;
|
|
|
|
};
|
|
|
|
|
|
|
|
size_t num_of_keys = keys.size();
|
|
|
|
assert(values.size() == num_of_keys);
|
|
|
|
assert(statuses.size() == num_of_keys);
|
|
|
|
for (size_t i = 0; i < num_of_keys; ++i) {
|
|
|
|
bool check_result = true;
|
|
|
|
if (use_txn) {
|
|
|
|
assert(ryw_expected_values.size() == num_of_keys);
|
|
|
|
check_result = check_multiget(keys[i], values[i], statuses[i],
|
|
|
|
ryw_expected_values[i]);
|
|
|
|
} else {
|
|
|
|
check_result = check_multiget(keys[i], values[i], statuses[i],
|
|
|
|
std::nullopt /* ryw_expected_value */);
|
|
|
|
}
|
|
|
|
if (!check_result) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (readoptionscopy.snapshot) {
|
|
|
|
db_->ReleaseSnapshot(readoptionscopy.snapshot);
|
|
|
|
}
|
|
|
|
if (use_txn) {
|
|
|
|
RollbackTxn(txn);
|
|
|
|
}
|
|
|
|
return statuses;
|
|
|
|
}
|
|
|
|
|
|
|
|
void TestGetEntity(ThreadState* thread, const ReadOptions& read_opts,
|
|
|
|
const std::vector<int>& rand_column_families,
|
|
|
|
const std::vector<int64_t>& rand_keys) override {
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
fault_fs_guard->EnableErrorInjection();
|
|
|
|
SharedState::ignore_read_error = false;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(thread);
|
|
|
|
|
|
|
|
SharedState* const shared = thread->shared;
|
|
|
|
assert(shared);
|
|
|
|
|
|
|
|
assert(!rand_column_families.empty());
|
|
|
|
assert(!rand_keys.empty());
|
|
|
|
|
|
|
|
std::unique_ptr<MutexLock> lock(new MutexLock(
|
|
|
|
shared->GetMutexForKey(rand_column_families[0], rand_keys[0])));
|
|
|
|
|
|
|
|
assert(rand_column_families[0] >= 0);
|
|
|
|
assert(rand_column_families[0] < static_cast<int>(column_families_.size()));
|
|
|
|
|
|
|
|
ColumnFamilyHandle* const cfh = column_families_[rand_column_families[0]];
|
|
|
|
assert(cfh);
|
|
|
|
|
|
|
|
const std::string key = Key(rand_keys[0]);
|
|
|
|
|
|
|
|
PinnableWideColumns from_db;
|
|
|
|
|
|
|
|
const Status s = db_->GetEntity(read_opts, cfh, key, &from_db);
|
|
|
|
|
|
|
|
int error_count = 0;
|
|
|
|
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
error_count = fault_fs_guard->GetAndResetErrorCount();
|
|
|
|
}
|
|
|
|
|
|
|
|
if (s.ok()) {
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
if (error_count && !SharedState::ignore_read_error) {
|
|
|
|
// Grab mutex so multiple threads don't try to print the
|
|
|
|
// stack trace at the same time
|
|
|
|
MutexLock l(shared->GetMutex());
|
|
|
|
fprintf(stderr, "Didn't get expected error from GetEntity\n");
|
|
|
|
fprintf(stderr, "Call stack that injected the fault\n");
|
|
|
|
fault_fs_guard->PrintFaultBacktrace();
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
thread->stats.AddGets(1, 1);
|
|
|
|
|
|
|
|
if (!FLAGS_skip_verifydb) {
|
|
|
|
const WideColumns& columns = from_db.columns();
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
ExpectedValue expected =
|
|
|
|
shared->Get(rand_column_families[0], rand_keys[0]);
|
|
|
|
if (!VerifyWideColumns(columns)) {
|
|
|
|
shared->SetVerificationFailure();
|
|
|
|
fprintf(stderr,
|
|
|
|
"error : inconsistent columns returned by GetEntity for key "
|
|
|
|
"%s: %s\n",
|
|
|
|
StringToHex(key).c_str(), WideColumnsToHex(columns).c_str());
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
} else if (ExpectedValueHelper::MustHaveNotExisted(expected,
|
|
|
|
expected)) {
|
|
|
|
shared->SetVerificationFailure();
|
|
|
|
fprintf(
|
|
|
|
stderr,
|
|
|
|
"error : inconsistent values for key %s: GetEntity returns %s, "
|
|
|
|
"expected state does not have the key.\n",
|
|
|
|
StringToHex(key).c_str(), WideColumnsToHex(columns).c_str());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else if (s.IsNotFound()) {
|
|
|
|
thread->stats.AddGets(1, 0);
|
|
|
|
|
|
|
|
if (!FLAGS_skip_verifydb) {
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
ExpectedValue expected =
|
|
|
|
shared->Get(rand_column_families[0], rand_keys[0]);
|
|
|
|
if (ExpectedValueHelper::MustHaveExisted(expected, expected)) {
|
|
|
|
shared->SetVerificationFailure();
|
|
|
|
fprintf(stderr,
|
|
|
|
"error : inconsistent values for key %s: expected state has "
|
|
|
|
"the key, GetEntity returns NotFound.\n",
|
|
|
|
StringToHex(key).c_str());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if (error_count == 0) {
|
|
|
|
thread->stats.AddErrors(1);
|
|
|
|
} else {
|
|
|
|
thread->stats.AddVerifiedErrors(1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
fault_fs_guard->DisableErrorInjection();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void TestMultiGetEntity(ThreadState* thread, const ReadOptions& read_opts,
|
|
|
|
const std::vector<int>& rand_column_families,
|
|
|
|
const std::vector<int64_t>& rand_keys) override {
|
|
|
|
assert(thread);
|
|
|
|
|
|
|
|
ManagedSnapshot snapshot_guard(db_);
|
|
|
|
|
|
|
|
ReadOptions read_opts_copy(read_opts);
|
|
|
|
read_opts_copy.snapshot = snapshot_guard.snapshot();
|
|
|
|
|
|
|
|
assert(!rand_column_families.empty());
|
|
|
|
assert(rand_column_families[0] >= 0);
|
|
|
|
assert(rand_column_families[0] < static_cast<int>(column_families_.size()));
|
|
|
|
|
|
|
|
ColumnFamilyHandle* const cfh = column_families_[rand_column_families[0]];
|
|
|
|
assert(cfh);
|
|
|
|
|
|
|
|
assert(!rand_keys.empty());
|
|
|
|
|
|
|
|
const size_t num_keys = rand_keys.size();
|
|
|
|
|
|
|
|
std::vector<std::string> keys(num_keys);
|
|
|
|
std::vector<Slice> key_slices(num_keys);
|
|
|
|
|
|
|
|
for (size_t i = 0; i < num_keys; ++i) {
|
|
|
|
keys[i] = Key(rand_keys[i]);
|
|
|
|
key_slices[i] = keys[i];
|
|
|
|
}
|
|
|
|
|
|
|
|
std::vector<PinnableWideColumns> results(num_keys);
|
|
|
|
std::vector<Status> statuses(num_keys);
|
|
|
|
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
fault_fs_guard->EnableErrorInjection();
|
|
|
|
SharedState::ignore_read_error = false;
|
|
|
|
}
|
|
|
|
|
|
|
|
db_->MultiGetEntity(read_opts_copy, cfh, num_keys, key_slices.data(),
|
|
|
|
results.data(), statuses.data());
|
|
|
|
|
|
|
|
int error_count = 0;
|
|
|
|
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
error_count = fault_fs_guard->GetAndResetErrorCount();
|
|
|
|
|
|
|
|
if (error_count && !SharedState::ignore_read_error) {
|
|
|
|
int stat_nok = 0;
|
|
|
|
for (const auto& s : statuses) {
|
|
|
|
if (!s.ok() && !s.IsNotFound()) {
|
|
|
|
stat_nok++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (stat_nok < error_count) {
|
|
|
|
// Grab mutex so multiple threads don't try to print the
|
|
|
|
// stack trace at the same time
|
|
|
|
assert(thread->shared);
|
|
|
|
MutexLock l(thread->shared->GetMutex());
|
|
|
|
|
|
|
|
fprintf(stderr, "Didn't get expected error from MultiGetEntity\n");
|
|
|
|
fprintf(stderr, "num_keys %zu Expected %d errors, seen %d\n",
|
|
|
|
num_keys, error_count, stat_nok);
|
|
|
|
fprintf(stderr, "Call stack that injected the fault\n");
|
|
|
|
fault_fs_guard->PrintFaultBacktrace();
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
fault_fs_guard->DisableErrorInjection();
|
|
|
|
}
|
|
|
|
|
|
|
|
const bool check_get_entity = !error_count && thread->rand.OneIn(4);
|
|
|
|
|
|
|
|
for (size_t i = 0; i < num_keys; ++i) {
|
|
|
|
const Status& s = statuses[i];
|
|
|
|
|
|
|
|
bool is_consistent = true;
|
|
|
|
|
|
|
|
if (s.ok() && !VerifyWideColumns(results[i].columns())) {
|
|
|
|
fprintf(
|
|
|
|
stderr,
|
|
|
|
"error : inconsistent columns returned by MultiGetEntity for key "
|
|
|
|
"%s: %s\n",
|
|
|
|
StringToHex(keys[i]).c_str(),
|
|
|
|
WideColumnsToHex(results[i].columns()).c_str());
|
|
|
|
is_consistent = false;
|
|
|
|
} else if (check_get_entity && (s.ok() || s.IsNotFound())) {
|
|
|
|
PinnableWideColumns cmp_result;
|
|
|
|
|
|
|
|
const Status cmp_s =
|
|
|
|
db_->GetEntity(read_opts_copy, cfh, key_slices[i], &cmp_result);
|
|
|
|
|
|
|
|
if (!cmp_s.ok() && !cmp_s.IsNotFound()) {
|
|
|
|
fprintf(stderr, "GetEntity error: %s\n", cmp_s.ToString().c_str());
|
|
|
|
is_consistent = false;
|
|
|
|
} else if (cmp_s.IsNotFound()) {
|
|
|
|
if (s.ok()) {
|
|
|
|
fprintf(stderr,
|
|
|
|
"Inconsistent results for key %s: MultiGetEntity returned "
|
|
|
|
"ok, GetEntity returned not found\n",
|
|
|
|
StringToHex(keys[i]).c_str());
|
|
|
|
is_consistent = false;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
assert(cmp_s.ok());
|
|
|
|
|
|
|
|
if (s.IsNotFound()) {
|
|
|
|
fprintf(stderr,
|
|
|
|
"Inconsistent results for key %s: MultiGetEntity returned "
|
|
|
|
"not found, GetEntity returned ok\n",
|
|
|
|
StringToHex(keys[i]).c_str());
|
|
|
|
is_consistent = false;
|
|
|
|
} else {
|
|
|
|
assert(s.ok());
|
|
|
|
|
|
|
|
if (results[i] != cmp_result) {
|
|
|
|
fprintf(
|
|
|
|
stderr,
|
|
|
|
"Inconsistent results for key %s: MultiGetEntity returned "
|
|
|
|
"%s, GetEntity returned %s\n",
|
|
|
|
StringToHex(keys[i]).c_str(),
|
|
|
|
WideColumnsToHex(results[i].columns()).c_str(),
|
|
|
|
WideColumnsToHex(cmp_result.columns()).c_str());
|
|
|
|
is_consistent = false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!is_consistent) {
|
|
|
|
fprintf(stderr,
|
|
|
|
"TestMultiGetEntity error: results are not consistent\n");
|
|
|
|
thread->stats.AddErrors(1);
|
|
|
|
// Fail fast to preserve the DB state
|
|
|
|
thread->shared->SetVerificationFailure();
|
|
|
|
break;
|
|
|
|
} else if (s.ok()) {
|
|
|
|
thread->stats.AddGets(1, 1);
|
|
|
|
} else if (s.IsNotFound()) {
|
|
|
|
thread->stats.AddGets(1, 0);
|
|
|
|
} else {
|
|
|
|
if (error_count == 0) {
|
|
|
|
fprintf(stderr, "MultiGetEntity error: %s\n", s.ToString().c_str());
|
|
|
|
thread->stats.AddErrors(1);
|
|
|
|
} else {
|
|
|
|
thread->stats.AddVerifiedErrors(1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
Status TestPrefixScan(ThreadState* thread, const ReadOptions& read_opts,
|
|
|
|
const std::vector<int>& rand_column_families,
|
|
|
|
const std::vector<int64_t>& rand_keys) override {
|
|
|
|
assert(!rand_column_families.empty());
|
|
|
|
assert(!rand_keys.empty());
|
|
|
|
|
|
|
|
ColumnFamilyHandle* const cfh = column_families_[rand_column_families[0]];
|
|
|
|
assert(cfh);
|
|
|
|
|
|
|
|
const std::string key = Key(rand_keys[0]);
|
|
|
|
const Slice prefix(key.data(), FLAGS_prefix_size);
|
|
|
|
|
|
|
|
std::string upper_bound;
|
|
|
|
Slice ub_slice;
|
|
|
|
ReadOptions ro_copy = read_opts;
|
|
|
|
|
|
|
|
// Get the next prefix first and then see if we want to set upper bound.
|
|
|
|
// We'll use the next prefix in an assertion later on
|
|
|
|
if (GetNextPrefix(prefix, &upper_bound) && thread->rand.OneIn(2)) {
|
|
|
|
// For half of the time, set the upper bound to the next prefix
|
|
|
|
ub_slice = Slice(upper_bound);
|
|
|
|
ro_copy.iterate_upper_bound = &ub_slice;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string read_ts_str;
|
|
|
|
Slice read_ts_slice;
|
|
|
|
MaybeUseOlderTimestampForRangeScan(thread, read_ts_str, read_ts_slice,
|
|
|
|
ro_copy);
|
|
|
|
|
|
|
|
std::unique_ptr<Iterator> iter(db_->NewIterator(ro_copy, cfh));
|
|
|
|
|
|
|
|
uint64_t count = 0;
|
|
|
|
Status s;
|
|
|
|
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
fault_fs_guard->EnableErrorInjection();
|
|
|
|
SharedState::ignore_read_error = false;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (iter->Seek(prefix); iter->Valid() && iter->key().starts_with(prefix);
|
|
|
|
iter->Next()) {
|
|
|
|
++count;
|
|
|
|
|
|
|
|
// When iter_start_ts is set, iterator exposes internal keys, including
|
|
|
|
// tombstones; however, we want to perform column validation only for
|
|
|
|
// value-like types.
|
|
|
|
if (ro_copy.iter_start_ts) {
|
|
|
|
const ValueType value_type = ExtractValueType(iter->key());
|
|
|
|
if (value_type != kTypeValue && value_type != kTypeBlobIndex &&
|
|
|
|
value_type != kTypeWideColumnEntity) {
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!VerifyWideColumns(iter->value(), iter->columns())) {
|
|
|
|
s = Status::Corruption("Value and columns inconsistent",
|
|
|
|
DebugString(iter->value(), iter->columns()));
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (ro_copy.iter_start_ts == nullptr) {
|
|
|
|
assert(count <= GetPrefixKeyCount(prefix.ToString(), upper_bound));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (s.ok()) {
|
|
|
|
s = iter->status();
|
|
|
|
}
|
|
|
|
|
|
|
|
uint64_t error_count = 0;
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
error_count = fault_fs_guard->GetAndResetErrorCount();
|
|
|
|
}
|
|
|
|
if (!s.ok() && (!fault_fs_guard || (fault_fs_guard && !error_count))) {
|
|
|
|
fprintf(stderr, "TestPrefixScan error: %s\n", s.ToString().c_str());
|
|
|
|
thread->stats.AddErrors(1);
|
|
|
|
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (fault_fs_guard) {
|
|
|
|
fault_fs_guard->DisableErrorInjection();
|
|
|
|
}
|
|
|
|
thread->stats.AddPrefixes(1, count);
|
|
|
|
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
Status TestPut(ThreadState* thread, WriteOptions& write_opts,
|
|
|
|
const ReadOptions& read_opts,
|
|
|
|
const std::vector<int>& rand_column_families,
|
|
|
|
const std::vector<int64_t>& rand_keys,
|
|
|
|
char (&value)[100]) override {
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
assert(!rand_column_families.empty());
|
|
|
|
assert(!rand_keys.empty());
|
|
|
|
|
|
|
|
auto shared = thread->shared;
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
assert(shared);
|
|
|
|
|
|
|
|
const int64_t max_key = shared->GetMaxKey();
|
|
|
|
|
|
|
|
int64_t rand_key = rand_keys[0];
|
|
|
|
int rand_column_family = rand_column_families[0];
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
std::string write_ts;
|
|
|
|
|
|
|
|
std::unique_ptr<MutexLock> lock(
|
|
|
|
new MutexLock(shared->GetMutexForKey(rand_column_family, rand_key)));
|
|
|
|
while (!shared->AllowsOverwrite(rand_key) &&
|
|
|
|
(FLAGS_use_merge || shared->Exists(rand_column_family, rand_key))) {
|
|
|
|
lock.reset();
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
|
|
|
|
rand_key = thread->rand.Next() % max_key;
|
|
|
|
rand_column_family = thread->rand.Next() % FLAGS_column_families;
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
|
|
|
|
lock.reset(
|
|
|
|
new MutexLock(shared->GetMutexForKey(rand_column_family, rand_key)));
|
Add user-defined timestamps to db_stress (#8061)
Summary:
Add some basic test for user-defined timestamp to db_stress. Currently,
read with timestamp always tries to read using the current timestamp.
Due to the per-key timestamp-sequence ordering constraint, we only add timestamp-
related tests to the `NonBatchedOpsStressTest` since this test serializes accesses
to the same key and uses a file to cross-check data correctness.
The timestamp feature is not supported in a number of components, e.g. Merge, SingleDelete,
DeleteRange, CompactionFilter, Readonly instance, secondary instance, SST file ingestion, transaction,
etc. Therefore, db_stress should exit if user enables both timestamp and these features at the same
time. The (currently) incompatible features can be found in
`CheckAndSetOptionsForUserTimestamp`.
This PR also fixes a bug triggered when timestamp is enabled together with
`index_type=kBinarySearchWithFirstKey`. This bug fix will also be in another separate PR
with more unit tests coverage. Fixing it here because I do not want to exclude the index type
from crash test.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/8061
Test Plan: make crash_test_with_ts
Reviewed By: jay-zhuang
Differential Revision: D27056282
Pulled By: riversand963
fbshipit-source-id: c3e00ad1023fdb9ebbdf9601ec18270c5e2925a9
4 years ago
|
|
|
if (FLAGS_user_timestamp_size > 0) {
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
write_ts = GetNowNanos();
|
Add user-defined timestamps to db_stress (#8061)
Summary:
Add some basic test for user-defined timestamp to db_stress. Currently,
read with timestamp always tries to read using the current timestamp.
Due to the per-key timestamp-sequence ordering constraint, we only add timestamp-
related tests to the `NonBatchedOpsStressTest` since this test serializes accesses
to the same key and uses a file to cross-check data correctness.
The timestamp feature is not supported in a number of components, e.g. Merge, SingleDelete,
DeleteRange, CompactionFilter, Readonly instance, secondary instance, SST file ingestion, transaction,
etc. Therefore, db_stress should exit if user enables both timestamp and these features at the same
time. The (currently) incompatible features can be found in
`CheckAndSetOptionsForUserTimestamp`.
This PR also fixes a bug triggered when timestamp is enabled together with
`index_type=kBinarySearchWithFirstKey`. This bug fix will also be in another separate PR
with more unit tests coverage. Fixing it here because I do not want to exclude the index type
from crash test.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/8061
Test Plan: make crash_test_with_ts
Reviewed By: jay-zhuang
Differential Revision: D27056282
Pulled By: riversand963
fbshipit-source-id: c3e00ad1023fdb9ebbdf9601ec18270c5e2925a9
4 years ago
|
|
|
}
|
|
|
|
}
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
|
|
|
|
if (write_ts.empty() && FLAGS_user_timestamp_size) {
|
|
|
|
write_ts = GetNowNanos();
|
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
|
|
|
}
|
|
|
|
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
const std::string k = Key(rand_key);
|
|
|
|
|
|
|
|
ColumnFamilyHandle* const cfh = column_families_[rand_column_family];
|
|
|
|
assert(cfh);
|
|
|
|
|
|
|
|
if (FLAGS_verify_before_write) {
|
|
|
|
std::string from_db;
|
|
|
|
Status s = db_->Get(read_opts, cfh, k, &from_db);
|
|
|
|
if (!VerifyOrSyncValue(rand_column_family, rand_key, read_opts, shared,
|
|
|
|
/* msg_prefix */ "Pre-Put Get verification",
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
from_db, s)) {
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
}
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
PendingExpectedValue pending_expected_value =
|
|
|
|
shared->PreparePut(rand_column_family, rand_key);
|
|
|
|
const uint32_t value_base = pending_expected_value.GetFinalValueBase();
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
const size_t sz = GenerateValue(value_base, value, sizeof(value));
|
|
|
|
const Slice v(value, sz);
|
|
|
|
|
|
|
|
|
|
|
|
Status s;
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
|
|
|
|
if (FLAGS_use_merge) {
|
|
|
|
if (!FLAGS_use_txn) {
|
|
|
|
if (FLAGS_user_timestamp_size == 0) {
|
|
|
|
s = db_->Merge(write_opts, cfh, k, v);
|
|
|
|
} else {
|
|
|
|
s = db_->Merge(write_opts, cfh, k, write_ts, v);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
Transaction* txn;
|
|
|
|
s = NewTxn(write_opts, &txn);
|
|
|
|
if (s.ok()) {
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
s = txn->Merge(cfh, k, v);
|
|
|
|
if (s.ok()) {
|
Snapshots with user-specified timestamps (#9879)
Summary:
In RocksDB, keys are associated with (internal) sequence numbers which denote when the keys are written
to the database. Sequence numbers in different RocksDB instances are unrelated, thus not comparable.
It is nice if we can associate sequence numbers with their corresponding actual timestamps. One thing we can
do is to support user-defined timestamp, which allows the applications to specify the format of custom timestamps
and encode a timestamp with each key. More details can be found at https://github.com/facebook/rocksdb/wiki/User-defined-Timestamp-%28Experimental%29.
This PR provides a different but complementary approach. We can associate rocksdb snapshots (defined in
https://github.com/facebook/rocksdb/blob/7.2.fb/include/rocksdb/snapshot.h#L20) with **user-specified** timestamps.
Since a snapshot is essentially an object representing a sequence number, this PR establishes a bi-directional mapping between sequence numbers and timestamps.
In the past, snapshots are usually taken by readers. The current super-version is grabbed, and a `rocksdb::Snapshot`
object is created with the last published sequence number of the super-version. You can see that the reader actually
has no good idea of what timestamp to assign to this snapshot, because by the time the `GetSnapshot()` is called,
an arbitrarily long period of time may have already elapsed since the last write, which is when the last published
sequence number is written.
This observation motivates the creation of "timestamped" snapshots on the write path. Currently, this functionality is
exposed only to the layer of `TransactionDB`. Application can tell RocksDB to create a snapshot when a transaction
commits, effectively associating the last sequence number with a timestamp. It is also assumed that application will
ensure any two snapshots with timestamps should satisfy the following:
```
snapshot1.seq < snapshot2.seq iff. snapshot1.ts < snapshot2.ts
```
If the application can guarantee that when a reader takes a timestamped snapshot, there is no active writes going on
in the database, then we also allow the user to use a new API `TransactionDB::CreateTimestampedSnapshot()` to create
a snapshot with associated timestamp.
Code example
```cpp
// Create a timestamped snapshot when committing transaction.
txn->SetCommitTimestamp(100);
txn->SetSnapshotOnNextOperation();
txn->Commit();
// A wrapper API for convenience
Status Transaction::CommitAndTryCreateSnapshot(
std::shared_ptr<TransactionNotifier> notifier,
TxnTimestamp ts,
std::shared_ptr<const Snapshot>* ret);
// Create a timestamped snapshot if caller guarantees no concurrent writes
std::pair<Status, std::shared_ptr<const Snapshot>> snapshot = txn_db->CreateTimestampedSnapshot(100);
```
The snapshots created in this way will be managed by RocksDB with ref-counting and potentially shared with
other readers. We provide the following APIs for readers to retrieve a snapshot given a timestamp.
```cpp
// Return the timestamped snapshot correponding to given timestamp. If ts is
// kMaxTxnTimestamp, then we return the latest timestamped snapshot if present.
// Othersise, we return the snapshot whose timestamp is equal to `ts`. If no
// such snapshot exists, then we return null.
std::shared_ptr<const Snapshot> TransactionDB::GetTimestampedSnapshot(TxnTimestamp ts) const;
// Return the latest timestamped snapshot if present.
std::shared_ptr<const Snapshot> TransactionDB::GetLatestTimestampedSnapshot() const;
```
We also provide two additional APIs for stats collection and reporting purposes.
```cpp
Status TransactionDB::GetAllTimestampedSnapshots(
std::vector<std::shared_ptr<const Snapshot>>& snapshots) const;
// Return timestamped snapshots whose timestamps fall in [ts_lb, ts_ub) and store them in `snapshots`.
Status TransactionDB::GetTimestampedSnapshots(
TxnTimestamp ts_lb,
TxnTimestamp ts_ub,
std::vector<std::shared_ptr<const Snapshot>>& snapshots) const;
```
To prevent the number of timestamped snapshots from growing infinitely, we provide the following API to release
timestamped snapshots whose timestamps are older than or equal to a given threshold.
```cpp
void TransactionDB::ReleaseTimestampedSnapshotsOlderThan(TxnTimestamp ts);
```
Before shutdown, RocksDB will release all timestamped snapshots.
Comparison with user-defined timestamp and how they can be combined:
User-defined timestamp persists every key with a timestamp, while timestamped snapshots maintain a volatile
mapping between snapshots (sequence numbers) and timestamps.
Different internal keys with the same user key but different timestamps will be treated as different by compaction,
thus a newer version will not hide older versions (with smaller timestamps) unless they are eligible for garbage collection.
In contrast, taking a timestamped snapshot at a certain sequence number and timestamp prevents all the keys visible in
this snapshot from been dropped by compaction. Here, visible means (seq < snapshot and most recent).
The timestamped snapshot supports the semantics of reading at an exact point in time.
Timestamped snapshots can also be used with user-defined timestamp.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9879
Test Plan:
```
make check
TEST_TMPDIR=/dev/shm make crash_test_with_txn
```
Reviewed By: siying
Differential Revision: D35783919
Pulled By: riversand963
fbshipit-source-id: 586ad905e169189e19d3bfc0cb0177a7239d1bd4
3 years ago
|
|
|
s = CommitTxn(txn, thread);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
} else if (FLAGS_use_put_entity_one_in > 0 &&
|
|
|
|
(value_base % FLAGS_use_put_entity_one_in) == 0) {
|
|
|
|
s = db_->PutEntity(write_opts, cfh, k,
|
|
|
|
GenerateWideColumns(value_base, v));
|
|
|
|
} else {
|
|
|
|
if (!FLAGS_use_txn) {
|
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
|
|
|
if (FLAGS_user_timestamp_size == 0) {
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
s = db_->Put(write_opts, cfh, k, v);
|
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
|
|
|
} else {
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
s = db_->Put(write_opts, cfh, k, write_ts, v);
|
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
|
|
|
}
|
|
|
|
} else {
|
|
|
|
Transaction* txn;
|
|
|
|
s = NewTxn(write_opts, &txn);
|
|
|
|
if (s.ok()) {
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
s = txn->Put(cfh, k, v);
|
|
|
|
if (s.ok()) {
|
Snapshots with user-specified timestamps (#9879)
Summary:
In RocksDB, keys are associated with (internal) sequence numbers which denote when the keys are written
to the database. Sequence numbers in different RocksDB instances are unrelated, thus not comparable.
It is nice if we can associate sequence numbers with their corresponding actual timestamps. One thing we can
do is to support user-defined timestamp, which allows the applications to specify the format of custom timestamps
and encode a timestamp with each key. More details can be found at https://github.com/facebook/rocksdb/wiki/User-defined-Timestamp-%28Experimental%29.
This PR provides a different but complementary approach. We can associate rocksdb snapshots (defined in
https://github.com/facebook/rocksdb/blob/7.2.fb/include/rocksdb/snapshot.h#L20) with **user-specified** timestamps.
Since a snapshot is essentially an object representing a sequence number, this PR establishes a bi-directional mapping between sequence numbers and timestamps.
In the past, snapshots are usually taken by readers. The current super-version is grabbed, and a `rocksdb::Snapshot`
object is created with the last published sequence number of the super-version. You can see that the reader actually
has no good idea of what timestamp to assign to this snapshot, because by the time the `GetSnapshot()` is called,
an arbitrarily long period of time may have already elapsed since the last write, which is when the last published
sequence number is written.
This observation motivates the creation of "timestamped" snapshots on the write path. Currently, this functionality is
exposed only to the layer of `TransactionDB`. Application can tell RocksDB to create a snapshot when a transaction
commits, effectively associating the last sequence number with a timestamp. It is also assumed that application will
ensure any two snapshots with timestamps should satisfy the following:
```
snapshot1.seq < snapshot2.seq iff. snapshot1.ts < snapshot2.ts
```
If the application can guarantee that when a reader takes a timestamped snapshot, there is no active writes going on
in the database, then we also allow the user to use a new API `TransactionDB::CreateTimestampedSnapshot()` to create
a snapshot with associated timestamp.
Code example
```cpp
// Create a timestamped snapshot when committing transaction.
txn->SetCommitTimestamp(100);
txn->SetSnapshotOnNextOperation();
txn->Commit();
// A wrapper API for convenience
Status Transaction::CommitAndTryCreateSnapshot(
std::shared_ptr<TransactionNotifier> notifier,
TxnTimestamp ts,
std::shared_ptr<const Snapshot>* ret);
// Create a timestamped snapshot if caller guarantees no concurrent writes
std::pair<Status, std::shared_ptr<const Snapshot>> snapshot = txn_db->CreateTimestampedSnapshot(100);
```
The snapshots created in this way will be managed by RocksDB with ref-counting and potentially shared with
other readers. We provide the following APIs for readers to retrieve a snapshot given a timestamp.
```cpp
// Return the timestamped snapshot correponding to given timestamp. If ts is
// kMaxTxnTimestamp, then we return the latest timestamped snapshot if present.
// Othersise, we return the snapshot whose timestamp is equal to `ts`. If no
// such snapshot exists, then we return null.
std::shared_ptr<const Snapshot> TransactionDB::GetTimestampedSnapshot(TxnTimestamp ts) const;
// Return the latest timestamped snapshot if present.
std::shared_ptr<const Snapshot> TransactionDB::GetLatestTimestampedSnapshot() const;
```
We also provide two additional APIs for stats collection and reporting purposes.
```cpp
Status TransactionDB::GetAllTimestampedSnapshots(
std::vector<std::shared_ptr<const Snapshot>>& snapshots) const;
// Return timestamped snapshots whose timestamps fall in [ts_lb, ts_ub) and store them in `snapshots`.
Status TransactionDB::GetTimestampedSnapshots(
TxnTimestamp ts_lb,
TxnTimestamp ts_ub,
std::vector<std::shared_ptr<const Snapshot>>& snapshots) const;
```
To prevent the number of timestamped snapshots from growing infinitely, we provide the following API to release
timestamped snapshots whose timestamps are older than or equal to a given threshold.
```cpp
void TransactionDB::ReleaseTimestampedSnapshotsOlderThan(TxnTimestamp ts);
```
Before shutdown, RocksDB will release all timestamped snapshots.
Comparison with user-defined timestamp and how they can be combined:
User-defined timestamp persists every key with a timestamp, while timestamped snapshots maintain a volatile
mapping between snapshots (sequence numbers) and timestamps.
Different internal keys with the same user key but different timestamps will be treated as different by compaction,
thus a newer version will not hide older versions (with smaller timestamps) unless they are eligible for garbage collection.
In contrast, taking a timestamped snapshot at a certain sequence number and timestamp prevents all the keys visible in
this snapshot from been dropped by compaction. Here, visible means (seq < snapshot and most recent).
The timestamped snapshot supports the semantics of reading at an exact point in time.
Timestamped snapshots can also be used with user-defined timestamp.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9879
Test Plan:
```
make check
TEST_TMPDIR=/dev/shm make crash_test_with_txn
```
Reviewed By: siying
Differential Revision: D35783919
Pulled By: riversand963
fbshipit-source-id: 586ad905e169189e19d3bfc0cb0177a7239d1bd4
3 years ago
|
|
|
s = CommitTxn(txn, thread);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
pending_expected_value.Commit();
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
|
|
|
|
if (!s.ok()) {
|
|
|
|
if (FLAGS_injest_error_severity >= 2) {
|
|
|
|
if (!is_db_stopped_ && s.severity() >= Status::Severity::kFatalError) {
|
|
|
|
is_db_stopped_ = true;
|
|
|
|
} else if (!is_db_stopped_ ||
|
|
|
|
s.severity() < Status::Severity::kFatalError) {
|
|
|
|
fprintf(stderr, "put or merge error: %s\n", s.ToString().c_str());
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
fprintf(stderr, "put or merge error: %s\n", s.ToString().c_str());
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
}
|
Add the PutEntity API to the stress/crash tests (#10760)
Summary:
The patch adds the `PutEntity` API to the non-batched, batched, and
CF consistency stress tests. Namely, when the new `db_stress` command
line parameter `use_put_entity_one_in` is greater than zero, one in
N writes on average is performed using `PutEntity` rather than `Put`.
The wide-column entity written has the generated value in its default
column; in addition, it contains up to three additional columns where
the original generated value is divided up between the column name and the
column value (with the column name containing the first k characters of
the generated value, and the column value containing the rest). Whether
`PutEntity` is used (and if so, how many columns the entity has) is completely
determined by the "value base" used to generate the value (that is, there is
no randomness involved). Assuming the same `use_put_entity_one_in` setting
is used across `db_stress` invocations, this enables us to reconstruct and
validate the entity during subsequent `db_stress` runs.
Note that `PutEntity` is currently incompatible with `Merge`, transactions, and
user-defined timestamps; these combinations are currently disabled/disallowed.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10760
Test Plan: Ran some batched, non-batched, and CF consistency stress tests using the script.
Reviewed By: riversand963
Differential Revision: D39939032
Pulled By: ltamasi
fbshipit-source-id: eafdf124e95993fb7d73158e3b006d11819f7fa9
2 years ago
|
|
|
|
|
|
|
thread->stats.AddBytesForWrites(1, sz);
|
|
|
|
PrintKeyValue(rand_column_family, static_cast<uint32_t>(rand_key), value,
|
|
|
|
sz);
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
Status TestDelete(ThreadState* thread, WriteOptions& write_opts,
|
|
|
|
const std::vector<int>& rand_column_families,
|
|
|
|
const std::vector<int64_t>& rand_keys) override {
|
|
|
|
int64_t rand_key = rand_keys[0];
|
|
|
|
int rand_column_family = rand_column_families[0];
|
|
|
|
auto shared = thread->shared;
|
|
|
|
|
|
|
|
std::unique_ptr<MutexLock> lock(
|
|
|
|
new MutexLock(shared->GetMutexForKey(rand_column_family, rand_key)));
|
|
|
|
|
|
|
|
// OPERATION delete
|
|
|
|
std::string write_ts_str = GetNowNanos();
|
|
|
|
Slice write_ts = write_ts_str;
|
|
|
|
|
|
|
|
std::string key_str = Key(rand_key);
|
|
|
|
Slice key = key_str;
|
|
|
|
auto cfh = column_families_[rand_column_family];
|
|
|
|
|
|
|
|
// Use delete if the key may be overwritten and a single deletion
|
|
|
|
// otherwise.
|
|
|
|
Status s;
|
|
|
|
if (shared->AllowsOverwrite(rand_key)) {
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
PendingExpectedValue pending_expected_value =
|
|
|
|
shared->PrepareDelete(rand_column_family, rand_key);
|
|
|
|
if (!FLAGS_use_txn) {
|
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
|
|
|
if (FLAGS_user_timestamp_size == 0) {
|
|
|
|
s = db_->Delete(write_opts, cfh, key);
|
|
|
|
} else {
|
|
|
|
s = db_->Delete(write_opts, cfh, key, write_ts);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
Transaction* txn;
|
|
|
|
s = NewTxn(write_opts, &txn);
|
|
|
|
if (s.ok()) {
|
|
|
|
s = txn->Delete(cfh, key);
|
|
|
|
if (s.ok()) {
|
Snapshots with user-specified timestamps (#9879)
Summary:
In RocksDB, keys are associated with (internal) sequence numbers which denote when the keys are written
to the database. Sequence numbers in different RocksDB instances are unrelated, thus not comparable.
It is nice if we can associate sequence numbers with their corresponding actual timestamps. One thing we can
do is to support user-defined timestamp, which allows the applications to specify the format of custom timestamps
and encode a timestamp with each key. More details can be found at https://github.com/facebook/rocksdb/wiki/User-defined-Timestamp-%28Experimental%29.
This PR provides a different but complementary approach. We can associate rocksdb snapshots (defined in
https://github.com/facebook/rocksdb/blob/7.2.fb/include/rocksdb/snapshot.h#L20) with **user-specified** timestamps.
Since a snapshot is essentially an object representing a sequence number, this PR establishes a bi-directional mapping between sequence numbers and timestamps.
In the past, snapshots are usually taken by readers. The current super-version is grabbed, and a `rocksdb::Snapshot`
object is created with the last published sequence number of the super-version. You can see that the reader actually
has no good idea of what timestamp to assign to this snapshot, because by the time the `GetSnapshot()` is called,
an arbitrarily long period of time may have already elapsed since the last write, which is when the last published
sequence number is written.
This observation motivates the creation of "timestamped" snapshots on the write path. Currently, this functionality is
exposed only to the layer of `TransactionDB`. Application can tell RocksDB to create a snapshot when a transaction
commits, effectively associating the last sequence number with a timestamp. It is also assumed that application will
ensure any two snapshots with timestamps should satisfy the following:
```
snapshot1.seq < snapshot2.seq iff. snapshot1.ts < snapshot2.ts
```
If the application can guarantee that when a reader takes a timestamped snapshot, there is no active writes going on
in the database, then we also allow the user to use a new API `TransactionDB::CreateTimestampedSnapshot()` to create
a snapshot with associated timestamp.
Code example
```cpp
// Create a timestamped snapshot when committing transaction.
txn->SetCommitTimestamp(100);
txn->SetSnapshotOnNextOperation();
txn->Commit();
// A wrapper API for convenience
Status Transaction::CommitAndTryCreateSnapshot(
std::shared_ptr<TransactionNotifier> notifier,
TxnTimestamp ts,
std::shared_ptr<const Snapshot>* ret);
// Create a timestamped snapshot if caller guarantees no concurrent writes
std::pair<Status, std::shared_ptr<const Snapshot>> snapshot = txn_db->CreateTimestampedSnapshot(100);
```
The snapshots created in this way will be managed by RocksDB with ref-counting and potentially shared with
other readers. We provide the following APIs for readers to retrieve a snapshot given a timestamp.
```cpp
// Return the timestamped snapshot correponding to given timestamp. If ts is
// kMaxTxnTimestamp, then we return the latest timestamped snapshot if present.
// Othersise, we return the snapshot whose timestamp is equal to `ts`. If no
// such snapshot exists, then we return null.
std::shared_ptr<const Snapshot> TransactionDB::GetTimestampedSnapshot(TxnTimestamp ts) const;
// Return the latest timestamped snapshot if present.
std::shared_ptr<const Snapshot> TransactionDB::GetLatestTimestampedSnapshot() const;
```
We also provide two additional APIs for stats collection and reporting purposes.
```cpp
Status TransactionDB::GetAllTimestampedSnapshots(
std::vector<std::shared_ptr<const Snapshot>>& snapshots) const;
// Return timestamped snapshots whose timestamps fall in [ts_lb, ts_ub) and store them in `snapshots`.
Status TransactionDB::GetTimestampedSnapshots(
TxnTimestamp ts_lb,
TxnTimestamp ts_ub,
std::vector<std::shared_ptr<const Snapshot>>& snapshots) const;
```
To prevent the number of timestamped snapshots from growing infinitely, we provide the following API to release
timestamped snapshots whose timestamps are older than or equal to a given threshold.
```cpp
void TransactionDB::ReleaseTimestampedSnapshotsOlderThan(TxnTimestamp ts);
```
Before shutdown, RocksDB will release all timestamped snapshots.
Comparison with user-defined timestamp and how they can be combined:
User-defined timestamp persists every key with a timestamp, while timestamped snapshots maintain a volatile
mapping between snapshots (sequence numbers) and timestamps.
Different internal keys with the same user key but different timestamps will be treated as different by compaction,
thus a newer version will not hide older versions (with smaller timestamps) unless they are eligible for garbage collection.
In contrast, taking a timestamped snapshot at a certain sequence number and timestamp prevents all the keys visible in
this snapshot from been dropped by compaction. Here, visible means (seq < snapshot and most recent).
The timestamped snapshot supports the semantics of reading at an exact point in time.
Timestamped snapshots can also be used with user-defined timestamp.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9879
Test Plan:
```
make check
TEST_TMPDIR=/dev/shm make crash_test_with_txn
```
Reviewed By: siying
Differential Revision: D35783919
Pulled By: riversand963
fbshipit-source-id: 586ad905e169189e19d3bfc0cb0177a7239d1bd4
3 years ago
|
|
|
s = CommitTxn(txn, thread);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
pending_expected_value.Commit();
|
|
|
|
|
|
|
|
thread->stats.AddDeletes(1);
|
|
|
|
if (!s.ok()) {
|
|
|
|
if (FLAGS_injest_error_severity >= 2) {
|
|
|
|
if (!is_db_stopped_ &&
|
|
|
|
s.severity() >= Status::Severity::kFatalError) {
|
|
|
|
is_db_stopped_ = true;
|
|
|
|
} else if (!is_db_stopped_ ||
|
|
|
|
s.severity() < Status::Severity::kFatalError) {
|
|
|
|
fprintf(stderr, "delete error: %s\n", s.ToString().c_str());
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
fprintf(stderr, "delete error: %s\n", s.ToString().c_str());
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else {
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
PendingExpectedValue pending_expected_value =
|
|
|
|
shared->PrepareSingleDelete(rand_column_family, rand_key);
|
|
|
|
if (!FLAGS_use_txn) {
|
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
|
|
|
if (FLAGS_user_timestamp_size == 0) {
|
|
|
|
s = db_->SingleDelete(write_opts, cfh, key);
|
|
|
|
} else {
|
|
|
|
s = db_->SingleDelete(write_opts, cfh, key, write_ts);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
Transaction* txn;
|
|
|
|
s = NewTxn(write_opts, &txn);
|
|
|
|
if (s.ok()) {
|
|
|
|
s = txn->SingleDelete(cfh, key);
|
|
|
|
if (s.ok()) {
|
Snapshots with user-specified timestamps (#9879)
Summary:
In RocksDB, keys are associated with (internal) sequence numbers which denote when the keys are written
to the database. Sequence numbers in different RocksDB instances are unrelated, thus not comparable.
It is nice if we can associate sequence numbers with their corresponding actual timestamps. One thing we can
do is to support user-defined timestamp, which allows the applications to specify the format of custom timestamps
and encode a timestamp with each key. More details can be found at https://github.com/facebook/rocksdb/wiki/User-defined-Timestamp-%28Experimental%29.
This PR provides a different but complementary approach. We can associate rocksdb snapshots (defined in
https://github.com/facebook/rocksdb/blob/7.2.fb/include/rocksdb/snapshot.h#L20) with **user-specified** timestamps.
Since a snapshot is essentially an object representing a sequence number, this PR establishes a bi-directional mapping between sequence numbers and timestamps.
In the past, snapshots are usually taken by readers. The current super-version is grabbed, and a `rocksdb::Snapshot`
object is created with the last published sequence number of the super-version. You can see that the reader actually
has no good idea of what timestamp to assign to this snapshot, because by the time the `GetSnapshot()` is called,
an arbitrarily long period of time may have already elapsed since the last write, which is when the last published
sequence number is written.
This observation motivates the creation of "timestamped" snapshots on the write path. Currently, this functionality is
exposed only to the layer of `TransactionDB`. Application can tell RocksDB to create a snapshot when a transaction
commits, effectively associating the last sequence number with a timestamp. It is also assumed that application will
ensure any two snapshots with timestamps should satisfy the following:
```
snapshot1.seq < snapshot2.seq iff. snapshot1.ts < snapshot2.ts
```
If the application can guarantee that when a reader takes a timestamped snapshot, there is no active writes going on
in the database, then we also allow the user to use a new API `TransactionDB::CreateTimestampedSnapshot()` to create
a snapshot with associated timestamp.
Code example
```cpp
// Create a timestamped snapshot when committing transaction.
txn->SetCommitTimestamp(100);
txn->SetSnapshotOnNextOperation();
txn->Commit();
// A wrapper API for convenience
Status Transaction::CommitAndTryCreateSnapshot(
std::shared_ptr<TransactionNotifier> notifier,
TxnTimestamp ts,
std::shared_ptr<const Snapshot>* ret);
// Create a timestamped snapshot if caller guarantees no concurrent writes
std::pair<Status, std::shared_ptr<const Snapshot>> snapshot = txn_db->CreateTimestampedSnapshot(100);
```
The snapshots created in this way will be managed by RocksDB with ref-counting and potentially shared with
other readers. We provide the following APIs for readers to retrieve a snapshot given a timestamp.
```cpp
// Return the timestamped snapshot correponding to given timestamp. If ts is
// kMaxTxnTimestamp, then we return the latest timestamped snapshot if present.
// Othersise, we return the snapshot whose timestamp is equal to `ts`. If no
// such snapshot exists, then we return null.
std::shared_ptr<const Snapshot> TransactionDB::GetTimestampedSnapshot(TxnTimestamp ts) const;
// Return the latest timestamped snapshot if present.
std::shared_ptr<const Snapshot> TransactionDB::GetLatestTimestampedSnapshot() const;
```
We also provide two additional APIs for stats collection and reporting purposes.
```cpp
Status TransactionDB::GetAllTimestampedSnapshots(
std::vector<std::shared_ptr<const Snapshot>>& snapshots) const;
// Return timestamped snapshots whose timestamps fall in [ts_lb, ts_ub) and store them in `snapshots`.
Status TransactionDB::GetTimestampedSnapshots(
TxnTimestamp ts_lb,
TxnTimestamp ts_ub,
std::vector<std::shared_ptr<const Snapshot>>& snapshots) const;
```
To prevent the number of timestamped snapshots from growing infinitely, we provide the following API to release
timestamped snapshots whose timestamps are older than or equal to a given threshold.
```cpp
void TransactionDB::ReleaseTimestampedSnapshotsOlderThan(TxnTimestamp ts);
```
Before shutdown, RocksDB will release all timestamped snapshots.
Comparison with user-defined timestamp and how they can be combined:
User-defined timestamp persists every key with a timestamp, while timestamped snapshots maintain a volatile
mapping between snapshots (sequence numbers) and timestamps.
Different internal keys with the same user key but different timestamps will be treated as different by compaction,
thus a newer version will not hide older versions (with smaller timestamps) unless they are eligible for garbage collection.
In contrast, taking a timestamped snapshot at a certain sequence number and timestamp prevents all the keys visible in
this snapshot from been dropped by compaction. Here, visible means (seq < snapshot and most recent).
The timestamped snapshot supports the semantics of reading at an exact point in time.
Timestamped snapshots can also be used with user-defined timestamp.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/9879
Test Plan:
```
make check
TEST_TMPDIR=/dev/shm make crash_test_with_txn
```
Reviewed By: siying
Differential Revision: D35783919
Pulled By: riversand963
fbshipit-source-id: 586ad905e169189e19d3bfc0cb0177a7239d1bd4
3 years ago
|
|
|
s = CommitTxn(txn, thread);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
pending_expected_value.Commit();
|
|
|
|
thread->stats.AddSingleDeletes(1);
|
|
|
|
if (!s.ok()) {
|
|
|
|
if (FLAGS_injest_error_severity >= 2) {
|
|
|
|
if (!is_db_stopped_ &&
|
|
|
|
s.severity() >= Status::Severity::kFatalError) {
|
|
|
|
is_db_stopped_ = true;
|
|
|
|
} else if (!is_db_stopped_ ||
|
|
|
|
s.severity() < Status::Severity::kFatalError) {
|
|
|
|
fprintf(stderr, "single delete error: %s\n", s.ToString().c_str());
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
fprintf(stderr, "single delete error: %s\n", s.ToString().c_str());
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
Status TestDeleteRange(ThreadState* thread, WriteOptions& write_opts,
|
|
|
|
const std::vector<int>& rand_column_families,
|
|
|
|
const std::vector<int64_t>& rand_keys) override {
|
|
|
|
// OPERATION delete range
|
|
|
|
std::vector<std::unique_ptr<MutexLock>> range_locks;
|
|
|
|
// delete range does not respect disallowed overwrites. the keys for
|
|
|
|
// which overwrites are disallowed are randomly distributed so it
|
|
|
|
// could be expensive to find a range where each key allows
|
|
|
|
// overwrites.
|
|
|
|
int64_t rand_key = rand_keys[0];
|
|
|
|
int rand_column_family = rand_column_families[0];
|
|
|
|
auto shared = thread->shared;
|
|
|
|
int64_t max_key = shared->GetMaxKey();
|
|
|
|
if (rand_key > max_key - FLAGS_range_deletion_width) {
|
|
|
|
rand_key =
|
|
|
|
thread->rand.Next() % (max_key - FLAGS_range_deletion_width + 1);
|
|
|
|
}
|
|
|
|
for (int j = 0; j < FLAGS_range_deletion_width; ++j) {
|
|
|
|
if (j == 0 ||
|
|
|
|
((rand_key + j) & ((1 << FLAGS_log2_keys_per_lock) - 1)) == 0) {
|
|
|
|
range_locks.emplace_back(new MutexLock(
|
|
|
|
shared->GetMutexForKey(rand_column_family, rand_key + j)));
|
|
|
|
}
|
|
|
|
}
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
std::vector<PendingExpectedValue> pending_expected_values =
|
|
|
|
shared->PrepareDeleteRange(rand_column_family, rand_key,
|
|
|
|
rand_key + FLAGS_range_deletion_width);
|
|
|
|
const int covered = static_cast<int>(pending_expected_values.size());
|
|
|
|
std::string keystr = Key(rand_key);
|
|
|
|
Slice key = keystr;
|
|
|
|
auto cfh = column_families_[rand_column_family];
|
|
|
|
std::string end_keystr = Key(rand_key + FLAGS_range_deletion_width);
|
|
|
|
Slice end_key = end_keystr;
|
|
|
|
std::string write_ts_str;
|
|
|
|
Slice write_ts;
|
|
|
|
Status s;
|
|
|
|
if (FLAGS_user_timestamp_size) {
|
|
|
|
write_ts_str = GetNowNanos();
|
|
|
|
write_ts = write_ts_str;
|
|
|
|
s = db_->DeleteRange(write_opts, cfh, key, end_key, write_ts);
|
|
|
|
} else {
|
|
|
|
s = db_->DeleteRange(write_opts, cfh, key, end_key);
|
|
|
|
}
|
|
|
|
if (!s.ok()) {
|
|
|
|
if (FLAGS_injest_error_severity >= 2) {
|
|
|
|
if (!is_db_stopped_ && s.severity() >= Status::Severity::kFatalError) {
|
|
|
|
is_db_stopped_ = true;
|
|
|
|
} else if (!is_db_stopped_ ||
|
|
|
|
s.severity() < Status::Severity::kFatalError) {
|
|
|
|
fprintf(stderr, "delete range error: %s\n", s.ToString().c_str());
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
fprintf(stderr, "delete range error: %s\n", s.ToString().c_str());
|
|
|
|
std::terminate();
|
|
|
|
}
|
|
|
|
}
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
for (PendingExpectedValue& pending_expected_value :
|
|
|
|
pending_expected_values) {
|
|
|
|
pending_expected_value.Commit();
|
|
|
|
}
|
|
|
|
thread->stats.AddRangeDeletions(1);
|
|
|
|
thread->stats.AddCoveredByRangeDeletions(covered);
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
void TestIngestExternalFile(ThreadState* thread,
|
|
|
|
const std::vector<int>& rand_column_families,
|
|
|
|
const std::vector<int64_t>& rand_keys) override {
|
|
|
|
const std::string sst_filename =
|
|
|
|
FLAGS_db + "/." + std::to_string(thread->tid) + ".sst";
|
|
|
|
Status s;
|
|
|
|
if (db_stress_env->FileExists(sst_filename).ok()) {
|
|
|
|
// Maybe we terminated abnormally before, so cleanup to give this file
|
|
|
|
// ingestion a clean slate
|
|
|
|
s = db_stress_env->DeleteFile(sst_filename);
|
|
|
|
}
|
|
|
|
|
|
|
|
SstFileWriter sst_file_writer(EnvOptions(options_), options_);
|
|
|
|
if (s.ok()) {
|
|
|
|
s = sst_file_writer.Open(sst_filename);
|
|
|
|
}
|
|
|
|
int64_t key_base = rand_keys[0];
|
|
|
|
int column_family = rand_column_families[0];
|
|
|
|
std::vector<std::unique_ptr<MutexLock>> range_locks;
|
|
|
|
range_locks.reserve(FLAGS_ingest_external_file_width);
|
|
|
|
std::vector<int64_t> keys;
|
|
|
|
keys.reserve(FLAGS_ingest_external_file_width);
|
|
|
|
std::vector<uint32_t> values;
|
|
|
|
values.reserve(FLAGS_ingest_external_file_width);
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
std::vector<PendingExpectedValue> pending_expected_values;
|
|
|
|
pending_expected_values.reserve(FLAGS_ingest_external_file_width);
|
|
|
|
SharedState* shared = thread->shared;
|
|
|
|
|
|
|
|
assert(FLAGS_nooverwritepercent < 100);
|
|
|
|
// Grab locks, set pending state on expected values, and add keys
|
|
|
|
for (int64_t key = key_base;
|
|
|
|
s.ok() && key < shared->GetMaxKey() &&
|
|
|
|
static_cast<int32_t>(keys.size()) < FLAGS_ingest_external_file_width;
|
|
|
|
++key) {
|
|
|
|
if (key == key_base ||
|
|
|
|
(key & ((1 << FLAGS_log2_keys_per_lock) - 1)) == 0) {
|
|
|
|
range_locks.emplace_back(
|
|
|
|
new MutexLock(shared->GetMutexForKey(column_family, key)));
|
|
|
|
}
|
|
|
|
if (!shared->AllowsOverwrite(key)) {
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
// We could alternatively include `key` that is deleted.
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
keys.push_back(key);
|
|
|
|
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
PendingExpectedValue pending_expected_value =
|
|
|
|
shared->PreparePut(column_family, key);
|
|
|
|
const uint32_t value_base = pending_expected_value.GetFinalValueBase();
|
|
|
|
values.push_back(value_base);
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
pending_expected_values.push_back(pending_expected_value);
|
|
|
|
|
|
|
|
char value[100];
|
|
|
|
size_t value_len = GenerateValue(value_base, value, sizeof(value));
|
|
|
|
auto key_str = Key(key);
|
|
|
|
s = sst_file_writer.Put(Slice(key_str), Slice(value, value_len));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (s.ok() && keys.empty()) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (s.ok()) {
|
|
|
|
s = sst_file_writer.Finish();
|
|
|
|
}
|
|
|
|
if (s.ok()) {
|
|
|
|
s = db_->IngestExternalFile(column_families_[column_family],
|
|
|
|
{sst_filename}, IngestExternalFileOptions());
|
|
|
|
}
|
|
|
|
if (!s.ok()) {
|
|
|
|
fprintf(stderr, "file ingestion error: %s\n", s.ToString().c_str());
|
|
|
|
std::terminate();
|
|
|
|
}
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
|
|
|
|
for (size_t i = 0; i < pending_expected_values.size(); ++i) {
|
|
|
|
pending_expected_values[i].Commit();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
// Given a key K, this creates an iterator which scans the range
|
|
|
|
// [K, K + FLAGS_num_iterations) forward and backward.
|
|
|
|
// Then does a random sequence of Next/Prev operations.
|
|
|
|
Status TestIterateAgainstExpected(
|
|
|
|
ThreadState* thread, const ReadOptions& read_opts,
|
|
|
|
const std::vector<int>& rand_column_families,
|
|
|
|
const std::vector<int64_t>& rand_keys) override {
|
|
|
|
assert(thread);
|
|
|
|
assert(!rand_column_families.empty());
|
|
|
|
assert(!rand_keys.empty());
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
auto shared = thread->shared;
|
|
|
|
assert(shared);
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
int64_t max_key = shared->GetMaxKey();
|
|
|
|
|
|
|
|
const int64_t num_iter = static_cast<int64_t>(FLAGS_num_iterations);
|
|
|
|
|
|
|
|
int64_t lb = rand_keys[0];
|
|
|
|
if (lb > max_key - num_iter) {
|
|
|
|
lb = thread->rand.Next() % (max_key - num_iter + 1);
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
}
|
|
|
|
|
|
|
|
const int64_t ub = lb + num_iter;
|
|
|
|
|
|
|
|
// Lock the whole range over which we might iterate to ensure it doesn't
|
|
|
|
// change under us.
|
|
|
|
const int rand_column_family = rand_column_families[0];
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
|
|
|
|
// Testing parallel read and write to the same key with user timestamp
|
|
|
|
// is not currently supported
|
|
|
|
std::vector<std::unique_ptr<MutexLock>> range_locks;
|
|
|
|
if (FLAGS_user_timestamp_size > 0) {
|
|
|
|
range_locks = shared->GetLocksForKeyRange(rand_column_family, lb, ub);
|
|
|
|
}
|
|
|
|
|
|
|
|
ReadOptions ro(read_opts);
|
|
|
|
ro.total_order_seek = true;
|
|
|
|
|
Verify Iterator/Get() against expected state in only `no_batched_ops_test` (#10590)
Summary:
https://github.com/facebook/rocksdb/issues/10538 added `TestIterateAgainstExpected()` in `no_batched_ops_test` to verify iterator correctness against the in memory expected state. It is not compatible when run after some other stress tests, e.g. `TestPut()` in `batched_op_stress`, that either do not set expected state when writing to DB or use keys that cannot be parsed by `GetIntVal()`. The assert [here](https://github.com/facebook/rocksdb/blob/d17be55aab80b856f96f4af89f8d18fef96646b4/db_stress_tool/db_stress_common.h#L520) could fail. This PR fixed this issue by setting iterator upperbound to `max_key` when `destroy_db_initially=0` to avoid the key space that `batched_op_stress` touches.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10590
Test Plan:
```
# set up DB with batched_op_stress
./db_stress --test_batches_snapshots=1 --verify_iterator_with_expected_state_one_in=1 --max_key_len=3 --max_key=100000000 --skip_verifydb=1 --continuous_verification_interval=0 --writepercent=85 --delpercent=3 --delrangepercent=0 --iterpercent=10 --nooverwritepercent=1 --prefixpercent=0 --readpercent=2 --key_len_percent_dist=1,30,69
# Before this PR, the following test will fail the asserts with error msg like the following
# Assertion failed: (size_key <= key_gen_ctx.weights.size() * sizeof(uint64_t)), function GetIntVal, file db_stress_common.h, line 524.
./db_stress --verify_iterator_with_expected_state_one_in=1 --max_key_len=3 --max_key=100000000 --skip_verifydb=1 --continuous_verification_interval=0 --writepercent=0 --delpercent=3 --delrangepercent=0 --iterpercent=95 --nooverwritepercent=1 --prefixpercent=0 --readpercent=2 --key_len_percent_dist=1,30,69 --destroy_db_initially=0
```
Reviewed By: ajkr
Differential Revision: D39085243
Pulled By: cbi42
fbshipit-source-id: a7dfee2320c330773b623b442d730fd014ec7056
2 years ago
|
|
|
std::string read_ts_str;
|
|
|
|
Slice read_ts;
|
|
|
|
if (FLAGS_user_timestamp_size > 0) {
|
|
|
|
read_ts_str = GetNowNanos();
|
|
|
|
read_ts = read_ts_str;
|
|
|
|
ro.timestamp = &read_ts;
|
Verify Iterator/Get() against expected state in only `no_batched_ops_test` (#10590)
Summary:
https://github.com/facebook/rocksdb/issues/10538 added `TestIterateAgainstExpected()` in `no_batched_ops_test` to verify iterator correctness against the in memory expected state. It is not compatible when run after some other stress tests, e.g. `TestPut()` in `batched_op_stress`, that either do not set expected state when writing to DB or use keys that cannot be parsed by `GetIntVal()`. The assert [here](https://github.com/facebook/rocksdb/blob/d17be55aab80b856f96f4af89f8d18fef96646b4/db_stress_tool/db_stress_common.h#L520) could fail. This PR fixed this issue by setting iterator upperbound to `max_key` when `destroy_db_initially=0` to avoid the key space that `batched_op_stress` touches.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10590
Test Plan:
```
# set up DB with batched_op_stress
./db_stress --test_batches_snapshots=1 --verify_iterator_with_expected_state_one_in=1 --max_key_len=3 --max_key=100000000 --skip_verifydb=1 --continuous_verification_interval=0 --writepercent=85 --delpercent=3 --delrangepercent=0 --iterpercent=10 --nooverwritepercent=1 --prefixpercent=0 --readpercent=2 --key_len_percent_dist=1,30,69
# Before this PR, the following test will fail the asserts with error msg like the following
# Assertion failed: (size_key <= key_gen_ctx.weights.size() * sizeof(uint64_t)), function GetIntVal, file db_stress_common.h, line 524.
./db_stress --verify_iterator_with_expected_state_one_in=1 --max_key_len=3 --max_key=100000000 --skip_verifydb=1 --continuous_verification_interval=0 --writepercent=0 --delpercent=3 --delrangepercent=0 --iterpercent=95 --nooverwritepercent=1 --prefixpercent=0 --readpercent=2 --key_len_percent_dist=1,30,69 --destroy_db_initially=0
```
Reviewed By: ajkr
Differential Revision: D39085243
Pulled By: cbi42
fbshipit-source-id: a7dfee2320c330773b623b442d730fd014ec7056
2 years ago
|
|
|
}
|
|
|
|
|
Verify Iterator/Get() against expected state in only `no_batched_ops_test` (#10590)
Summary:
https://github.com/facebook/rocksdb/issues/10538 added `TestIterateAgainstExpected()` in `no_batched_ops_test` to verify iterator correctness against the in memory expected state. It is not compatible when run after some other stress tests, e.g. `TestPut()` in `batched_op_stress`, that either do not set expected state when writing to DB or use keys that cannot be parsed by `GetIntVal()`. The assert [here](https://github.com/facebook/rocksdb/blob/d17be55aab80b856f96f4af89f8d18fef96646b4/db_stress_tool/db_stress_common.h#L520) could fail. This PR fixed this issue by setting iterator upperbound to `max_key` when `destroy_db_initially=0` to avoid the key space that `batched_op_stress` touches.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10590
Test Plan:
```
# set up DB with batched_op_stress
./db_stress --test_batches_snapshots=1 --verify_iterator_with_expected_state_one_in=1 --max_key_len=3 --max_key=100000000 --skip_verifydb=1 --continuous_verification_interval=0 --writepercent=85 --delpercent=3 --delrangepercent=0 --iterpercent=10 --nooverwritepercent=1 --prefixpercent=0 --readpercent=2 --key_len_percent_dist=1,30,69
# Before this PR, the following test will fail the asserts with error msg like the following
# Assertion failed: (size_key <= key_gen_ctx.weights.size() * sizeof(uint64_t)), function GetIntVal, file db_stress_common.h, line 524.
./db_stress --verify_iterator_with_expected_state_one_in=1 --max_key_len=3 --max_key=100000000 --skip_verifydb=1 --continuous_verification_interval=0 --writepercent=0 --delpercent=3 --delrangepercent=0 --iterpercent=95 --nooverwritepercent=1 --prefixpercent=0 --readpercent=2 --key_len_percent_dist=1,30,69 --destroy_db_initially=0
```
Reviewed By: ajkr
Differential Revision: D39085243
Pulled By: cbi42
fbshipit-source-id: a7dfee2320c330773b623b442d730fd014ec7056
2 years ago
|
|
|
std::string max_key_str;
|
|
|
|
Slice max_key_slice;
|
|
|
|
if (!FLAGS_destroy_db_initially) {
|
|
|
|
max_key_str = Key(max_key);
|
|
|
|
max_key_slice = max_key_str;
|
|
|
|
// to restrict iterator from reading keys written in batched_op_stress
|
|
|
|
// that do not have expected state updated and may not be parseable by
|
|
|
|
// GetIntVal().
|
|
|
|
ro.iterate_upper_bound = &max_key_slice;
|
Verify Iterator/Get() against expected state in only `no_batched_ops_test` (#10590)
Summary:
https://github.com/facebook/rocksdb/issues/10538 added `TestIterateAgainstExpected()` in `no_batched_ops_test` to verify iterator correctness against the in memory expected state. It is not compatible when run after some other stress tests, e.g. `TestPut()` in `batched_op_stress`, that either do not set expected state when writing to DB or use keys that cannot be parsed by `GetIntVal()`. The assert [here](https://github.com/facebook/rocksdb/blob/d17be55aab80b856f96f4af89f8d18fef96646b4/db_stress_tool/db_stress_common.h#L520) could fail. This PR fixed this issue by setting iterator upperbound to `max_key` when `destroy_db_initially=0` to avoid the key space that `batched_op_stress` touches.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10590
Test Plan:
```
# set up DB with batched_op_stress
./db_stress --test_batches_snapshots=1 --verify_iterator_with_expected_state_one_in=1 --max_key_len=3 --max_key=100000000 --skip_verifydb=1 --continuous_verification_interval=0 --writepercent=85 --delpercent=3 --delrangepercent=0 --iterpercent=10 --nooverwritepercent=1 --prefixpercent=0 --readpercent=2 --key_len_percent_dist=1,30,69
# Before this PR, the following test will fail the asserts with error msg like the following
# Assertion failed: (size_key <= key_gen_ctx.weights.size() * sizeof(uint64_t)), function GetIntVal, file db_stress_common.h, line 524.
./db_stress --verify_iterator_with_expected_state_one_in=1 --max_key_len=3 --max_key=100000000 --skip_verifydb=1 --continuous_verification_interval=0 --writepercent=0 --delpercent=3 --delrangepercent=0 --iterpercent=95 --nooverwritepercent=1 --prefixpercent=0 --readpercent=2 --key_len_percent_dist=1,30,69 --destroy_db_initially=0
```
Reviewed By: ajkr
Differential Revision: D39085243
Pulled By: cbi42
fbshipit-source-id: a7dfee2320c330773b623b442d730fd014ec7056
2 years ago
|
|
|
}
|
|
|
|
|
|
|
|
ColumnFamilyHandle* const cfh = column_families_[rand_column_family];
|
|
|
|
assert(cfh);
|
|
|
|
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
const std::size_t expected_values_size = static_cast<std::size_t>(ub - lb);
|
|
|
|
std::vector<ExpectedValue> pre_read_expected_values;
|
|
|
|
std::vector<ExpectedValue> post_read_expected_values;
|
|
|
|
|
|
|
|
for (int64_t i = 0; i < static_cast<int64_t>(expected_values_size); ++i) {
|
|
|
|
pre_read_expected_values.push_back(
|
|
|
|
shared->Get(rand_column_family, i + lb));
|
|
|
|
}
|
|
|
|
std::unique_ptr<Iterator> iter(db_->NewIterator(ro, cfh));
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
for (int64_t i = 0; i < static_cast<int64_t>(expected_values_size); ++i) {
|
|
|
|
post_read_expected_values.push_back(
|
|
|
|
shared->Get(rand_column_family, i + lb));
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(pre_read_expected_values.size() == expected_values_size &&
|
|
|
|
pre_read_expected_values.size() == post_read_expected_values.size());
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
std::string op_logs;
|
|
|
|
|
|
|
|
auto check_columns = [&]() {
|
|
|
|
assert(iter);
|
|
|
|
assert(iter->Valid());
|
|
|
|
|
|
|
|
if (!VerifyWideColumns(iter->value(), iter->columns())) {
|
|
|
|
shared->SetVerificationFailure();
|
|
|
|
|
|
|
|
fprintf(stderr,
|
|
|
|
"Verification failed for key %s: "
|
|
|
|
"Value and columns inconsistent: value: %s, columns: %s\n",
|
|
|
|
Slice(iter->key()).ToString(/* hex */ true).c_str(),
|
|
|
|
iter->value().ToString(/* hex */ true).c_str(),
|
|
|
|
WideColumnsToHex(iter->columns()).c_str());
|
|
|
|
fprintf(stderr, "Column family: %s, op_logs: %s\n",
|
|
|
|
cfh->GetName().c_str(), op_logs.c_str());
|
|
|
|
|
|
|
|
thread->stats.AddErrors(1);
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
return true;
|
|
|
|
};
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
|
|
|
|
auto check_no_key_in_range = [&](int64_t start, int64_t end) {
|
|
|
|
for (auto j = std::max(start, lb); j < std::min(end, ub); ++j) {
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
std::size_t index = static_cast<std::size_t>(j - lb);
|
|
|
|
assert(index < pre_read_expected_values.size() &&
|
|
|
|
index < post_read_expected_values.size());
|
|
|
|
const ExpectedValue pre_read_expected_value =
|
|
|
|
pre_read_expected_values[index];
|
|
|
|
const ExpectedValue post_read_expected_value =
|
|
|
|
post_read_expected_values[index];
|
|
|
|
if (ExpectedValueHelper::MustHaveExisted(pre_read_expected_value,
|
|
|
|
post_read_expected_value)) {
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
// Fail fast to preserve the DB state.
|
|
|
|
thread->shared->SetVerificationFailure();
|
|
|
|
if (iter->Valid()) {
|
|
|
|
fprintf(stderr,
|
|
|
|
"Expected state has key %s, iterator is at key %s\n",
|
|
|
|
Slice(Key(j)).ToString(true).c_str(),
|
|
|
|
iter->key().ToString(true).c_str());
|
|
|
|
} else {
|
|
|
|
fprintf(stderr, "Expected state has key %s, iterator is invalid\n",
|
|
|
|
Slice(Key(j)).ToString(true).c_str());
|
|
|
|
}
|
|
|
|
fprintf(stderr, "Column family: %s, op_logs: %s\n",
|
|
|
|
cfh->GetName().c_str(), op_logs.c_str());
|
|
|
|
thread->stats.AddErrors(1);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return true;
|
|
|
|
};
|
|
|
|
|
|
|
|
// Forward and backward scan to ensure we cover the entire range [lb, ub).
|
|
|
|
// The random sequence Next and Prev test below tends to be very short
|
|
|
|
// ranged.
|
|
|
|
int64_t last_key = lb - 1;
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
std::string key_str = Key(lb);
|
|
|
|
iter->Seek(key_str);
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
op_logs += "S " + Slice(key_str).ToString(true) + " ";
|
|
|
|
|
|
|
|
uint64_t curr = 0;
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
while (true) {
|
|
|
|
if (!iter->Valid()) {
|
|
|
|
if (!iter->status().ok()) {
|
|
|
|
thread->shared->SetVerificationFailure();
|
|
|
|
fprintf(stderr, "TestIterate against expected state error: %s\n",
|
|
|
|
iter->status().ToString().c_str());
|
|
|
|
fprintf(stderr, "Column family: %s, op_logs: %s\n",
|
|
|
|
cfh->GetName().c_str(), op_logs.c_str());
|
|
|
|
thread->stats.AddErrors(1);
|
|
|
|
return iter->status();
|
|
|
|
}
|
|
|
|
if (!check_no_key_in_range(last_key + 1, ub)) {
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!check_columns()) {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
// iter is valid, the range (last_key, current key) was skipped
|
|
|
|
GetIntVal(iter->key().ToString(), &curr);
|
|
|
|
if (!check_no_key_in_range(last_key + 1, static_cast<int64_t>(curr))) {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
last_key = static_cast<int64_t>(curr);
|
|
|
|
if (last_key >= ub - 1) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
iter->Next();
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
op_logs += "N";
|
|
|
|
}
|
|
|
|
|
|
|
|
// backward scan
|
|
|
|
key_str = Key(ub - 1);
|
|
|
|
iter->SeekForPrev(key_str);
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
op_logs += " SFP " + Slice(key_str).ToString(true) + " ";
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
last_key = ub;
|
|
|
|
while (true) {
|
|
|
|
if (!iter->Valid()) {
|
|
|
|
if (!iter->status().ok()) {
|
|
|
|
thread->shared->SetVerificationFailure();
|
|
|
|
fprintf(stderr, "TestIterate against expected state error: %s\n",
|
|
|
|
iter->status().ToString().c_str());
|
|
|
|
fprintf(stderr, "Column family: %s, op_logs: %s\n",
|
|
|
|
cfh->GetName().c_str(), op_logs.c_str());
|
|
|
|
thread->stats.AddErrors(1);
|
|
|
|
return iter->status();
|
|
|
|
}
|
|
|
|
if (!check_no_key_in_range(lb, last_key)) {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!check_columns()) {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
// the range (current key, last key) was skipped
|
|
|
|
GetIntVal(iter->key().ToString(), &curr);
|
|
|
|
if (!check_no_key_in_range(static_cast<int64_t>(curr + 1), last_key)) {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
last_key = static_cast<int64_t>(curr);
|
|
|
|
if (last_key <= lb) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
iter->Prev();
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
op_logs += "P";
|
|
|
|
}
|
|
|
|
|
|
|
|
if (thread->rand.OneIn(2)) {
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
pre_read_expected_values.clear();
|
|
|
|
post_read_expected_values.clear();
|
|
|
|
// Refresh after forward/backward scan to allow higher chance of SV
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
// change.
|
|
|
|
for (int64_t i = 0; i < static_cast<int64_t>(expected_values_size); ++i) {
|
|
|
|
pre_read_expected_values.push_back(
|
|
|
|
shared->Get(rand_column_family, i + lb));
|
|
|
|
}
|
|
|
|
iter->Refresh();
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
for (int64_t i = 0; i < static_cast<int64_t>(expected_values_size); ++i) {
|
|
|
|
post_read_expected_values.push_back(
|
|
|
|
shared->Get(rand_column_family, i + lb));
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(pre_read_expected_values.size() == expected_values_size &&
|
|
|
|
pre_read_expected_values.size() ==
|
|
|
|
post_read_expected_values.size());
|
|
|
|
}
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
// start from middle of [lb, ub) otherwise it is easy to iterate out of
|
|
|
|
// locked range
|
|
|
|
const int64_t mid = lb + num_iter / 2;
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
key_str = Key(mid);
|
|
|
|
const Slice key(key_str);
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
if (thread->rand.OneIn(2)) {
|
|
|
|
iter->Seek(key);
|
|
|
|
op_logs += " S " + key.ToString(true) + " ";
|
|
|
|
if (!iter->Valid() && iter->status().ok()) {
|
|
|
|
if (!check_no_key_in_range(mid, ub)) {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
iter->SeekForPrev(key);
|
|
|
|
op_logs += " SFP " + key.ToString(true) + " ";
|
|
|
|
if (!iter->Valid() && iter->status().ok()) {
|
|
|
|
// iterator says nothing <= mid
|
|
|
|
if (!check_no_key_in_range(lb, mid + 1)) {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
for (int64_t i = 0; i < num_iter && iter->Valid(); ++i) {
|
|
|
|
if (!check_columns()) {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
GetIntVal(iter->key().ToString(), &curr);
|
|
|
|
if (static_cast<int64_t>(curr) < lb) {
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
iter->Next();
|
|
|
|
op_logs += "N";
|
|
|
|
} else if (static_cast<int64_t>(curr) >= ub) {
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
iter->Prev();
|
|
|
|
op_logs += "P";
|
|
|
|
} else {
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
const uint32_t value_base_from_db = GetValueBase(iter->value());
|
|
|
|
std::size_t index = static_cast<std::size_t>(curr - lb);
|
|
|
|
assert(index < pre_read_expected_values.size() &&
|
|
|
|
index < post_read_expected_values.size());
|
|
|
|
const ExpectedValue pre_read_expected_value =
|
|
|
|
pre_read_expected_values[index];
|
|
|
|
const ExpectedValue post_read_expected_value =
|
|
|
|
post_read_expected_values[index];
|
|
|
|
if (ExpectedValueHelper::MustHaveNotExisted(pre_read_expected_value,
|
|
|
|
post_read_expected_value) ||
|
|
|
|
!ExpectedValueHelper::InExpectedValueBaseRange(
|
|
|
|
value_base_from_db, pre_read_expected_value,
|
|
|
|
post_read_expected_value)) {
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
// Fail fast to preserve the DB state.
|
|
|
|
thread->shared->SetVerificationFailure();
|
|
|
|
fprintf(stderr, "Iterator has key %s, but expected state does not.\n",
|
|
|
|
iter->key().ToString(true).c_str());
|
|
|
|
fprintf(stderr, "Column family: %s, op_logs: %s\n",
|
|
|
|
cfh->GetName().c_str(), op_logs.c_str());
|
|
|
|
thread->stats.AddErrors(1);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
if (thread->rand.OneIn(2)) {
|
|
|
|
iter->Next();
|
|
|
|
op_logs += "N";
|
|
|
|
if (!iter->Valid()) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
uint64_t next = 0;
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
GetIntVal(iter->key().ToString(), &next);
|
|
|
|
if (!check_no_key_in_range(static_cast<int64_t>(curr + 1),
|
|
|
|
static_cast<int64_t>(next))) {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
iter->Prev();
|
|
|
|
op_logs += "P";
|
|
|
|
if (!iter->Valid()) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
uint64_t prev = 0;
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
GetIntVal(iter->key().ToString(), &prev);
|
|
|
|
if (!check_no_key_in_range(static_cast<int64_t>(prev + 1),
|
|
|
|
static_cast<int64_t>(curr))) {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
if (!iter->status().ok()) {
|
|
|
|
thread->shared->SetVerificationFailure();
|
|
|
|
fprintf(stderr, "TestIterate against expected state error: %s\n",
|
|
|
|
iter->status().ToString().c_str());
|
|
|
|
fprintf(stderr, "Column family: %s, op_logs: %s\n",
|
|
|
|
cfh->GetName().c_str(), op_logs.c_str());
|
|
|
|
thread->stats.AddErrors(1);
|
|
|
|
return iter->status();
|
|
|
|
}
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
thread->stats.AddIterations(1);
|
|
|
|
|
Add Iterator test against expected state to stress test (#10538)
Summary:
As mentioned in https://github.com/facebook/rocksdb/pull/5506#issuecomment-506021913,
`db_stress` does not have much verification for iterator correctness.
It has a `TestIterate()` function, but that is mainly for comparing results
between two iterators, one with `total_order_seek` and the other optionally
sets auto_prefix, upper/lower bounds. Commit 49a0581ad2462e31aa3f768afa769e0d33390f33
added a new `TestIterateAgainstExpected()` function that compares iterator against
expected state. It locks a range of keys, creates an iterator, does
a random sequence of `Next/Prev` and compares against expected state.
This PR is based on that commit, the main changes include some logs
(for easier debugging if a test fails), a forward and backward scan to
cover the entire locked key range, and a flag for optionally turning on
this version of Iterator testing.
Added constraint that the checks against expected state in
`TestIterateAgainstExpected()` and in `TestGet()` are only turned on
when `--skip_verifydb` flag is not set.
Remove the change log introduced in https://github.com/facebook/rocksdb/issues/10553.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10538
Test Plan:
Run `db_stress` with `--verify_iterator_with_expected_state_one_in=1`,
and a large `--iterpercent` and `--num_iterations`. Checked `op_logs`
manually to ensure expected coverage. Tweaked part of the code in
https://github.com/facebook/rocksdb/issues/10449 and stress test was able to catch it.
- internally run various flavor of crash test
Reviewed By: ajkr
Differential Revision: D38847269
Pulled By: cbi42
fbshipit-source-id: 8b4402a9bba9f6cfa08051943cd672579d489599
2 years ago
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
bool VerifyOrSyncValue(int cf, int64_t key, const ReadOptions& /*opts*/,
|
|
|
|
SharedState* shared, const std::string& value_from_db,
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
std::string msg_prefix, const Status& s) const {
|
|
|
|
if (shared->HasVerificationFailedYet()) {
|
|
|
|
return false;
|
|
|
|
}
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
const ExpectedValue expected_value = shared->Get(cf, key);
|
|
|
|
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
if (expected_value.PendingWrite() || expected_value.PendingDelete()) {
|
|
|
|
if (s.ok()) {
|
|
|
|
// Value exists in db, update state to reflect that
|
|
|
|
Slice slice(value_from_db);
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
uint32_t value_base = GetValueBase(slice);
|
|
|
|
shared->SyncPut(cf, key, value_base);
|
|
|
|
} else if (s.IsNotFound()) {
|
|
|
|
// Value doesn't exist in db, update state to reflect that
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
shared->SyncDelete(cf, key);
|
|
|
|
}
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
// compare value_from_db with the value in the shared state
|
|
|
|
if (s.ok()) {
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
const Slice slice(value_from_db);
|
|
|
|
const uint32_t value_base_from_db = GetValueBase(slice);
|
|
|
|
if (ExpectedValueHelper::MustHaveNotExisted(expected_value,
|
|
|
|
expected_value) ||
|
|
|
|
!ExpectedValueHelper::InExpectedValueBaseRange(
|
|
|
|
value_base_from_db, expected_value, expected_value)) {
|
|
|
|
VerificationAbort(shared, msg_prefix + ": Unexpected value found", cf,
|
|
|
|
key, value_from_db, "");
|
|
|
|
return false;
|
|
|
|
}
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
char expected_value_data[kValueMaxLen];
|
|
|
|
size_t expected_value_data_size =
|
|
|
|
GenerateValue(expected_value.GetValueBase(), expected_value_data,
|
|
|
|
sizeof(expected_value_data));
|
|
|
|
if (value_from_db.length() != expected_value_data_size) {
|
|
|
|
VerificationAbort(shared,
|
|
|
|
msg_prefix + ": Length of value read is not equal",
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
cf, key, value_from_db,
|
|
|
|
Slice(expected_value_data, expected_value_data_size));
|
|
|
|
return false;
|
|
|
|
}
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
if (memcmp(value_from_db.data(), expected_value_data,
|
|
|
|
expected_value_data_size) != 0) {
|
|
|
|
VerificationAbort(shared,
|
|
|
|
msg_prefix + ": Contents of value read don't match",
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
cf, key, value_from_db,
|
|
|
|
Slice(expected_value_data, expected_value_data_size));
|
|
|
|
return false;
|
|
|
|
}
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
} else if (s.IsNotFound()) {
|
|
|
|
if (ExpectedValueHelper::MustHaveExisted(expected_value,
|
|
|
|
expected_value)) {
|
|
|
|
char expected_value_data[kValueMaxLen];
|
|
|
|
size_t expected_value_data_size =
|
|
|
|
GenerateValue(expected_value.GetValueBase(), expected_value_data,
|
|
|
|
sizeof(expected_value_data));
|
|
|
|
VerificationAbort(
|
|
|
|
shared, msg_prefix + ": Value not found: " + s.ToString(), cf, key,
|
|
|
|
"", Slice(expected_value_data, expected_value_data_size));
|
|
|
|
return false;
|
|
|
|
}
|
Support parallel read and write/delete to same key in NonBatchedOpsStressTest (#11058)
Summary:
**Context:**
Current `NonBatchedOpsStressTest` does not allow multi-thread read (i.e, Get, Iterator) and write (i.e, Put, Merge) or delete to the same key. Every read or write/delete operation will acquire lock (`GetLocksForKeyRange`) on the target key to gain exclusive access to it. This does not align with RocksDB's nature of allowing multi-thread read and write/delete to the same key, that is concurrent threads can issue read/write/delete to RocksDB without external locking. Therefore this is a gap in our testing coverage.
To close the gap, biggest challenge remains in verifying db value against expected state in presence of parallel read and write/delete. The challenge is due to read/write/delete to the db and read/write to expected state is not within one atomic operation. Therefore we may not know the exact expected state of a certain db read, as by the time we read the expected state for that db read, another write to expected state for another db write to the same key might have changed the expected state.
**Summary:**
Credited to ajkr's idea, we now solve this challenge by breaking the 32-bits expected value of a key into different parts that can be read and write to in parallel.
Basically we divide the 32-bits expected value into `value_base` (corresponding to the previous whole 32 bits but now with some shrinking in the value base range we allow), `pending_write` (i.e, whether there is an ongoing concurrent write), `del_counter` (i.e, number of times a value has been deleted, analogous to value_base for write), `pending_delete` (similar to pending_write) and `deleted` (i.e whether a key is deleted).
Also, we need to use incremental `value_base` instead of random value base as before because we want to control the range of value base a correct db read result can possibly be in presence of parallel read and write. In that way, we can verify the correctness of the read against expected state more easily. This is at the cost of reducing the randomness of the value generated in NonBatchedOpsStressTest we are willing to accept.
(For detailed algorithm of how to use these parts to infer expected state of a key, see the PR)
Misc: hide value_base detail from callers of ExpectedState by abstracting related logics into ExpectedValue class
Pull Request resolved: https://github.com/facebook/rocksdb/pull/11058
Test Plan:
- Manual test of small number of keys (i.e, high chances of parallel read and write/delete to same key) with equally distributed read/write/deleted for 30 min
```
python3 tools/db_crashtest.py --simple {blackbox|whitebox} --sync_fault_injection=1 --skip_verifydb=0 --continuous_verification_interval=1000 --clear_column_family_one_in=0 --max_key=10 --column_families=1 --threads=32 --readpercent=25 --writepercent=25 --nooverwritepercent=0 --iterpercent=25 --verify_iterator_with_expected_state_one_in=1 --num_iterations=5 --delpercent=15 --delrangepercent=10 --range_deletion_width=5 --use_merge={0|1} --use_put_entity_one_in=0 --use_txn=0 --verify_before_write=0 --user_timestamp_size=0 --compact_files_one_in=1000 --compact_range_one_in=1000 --flush_one_in=1000 --get_property_one_in=1000 --ingest_external_file_one_in=100 --backup_one_in=100 --checkpoint_one_in=100 --approximate_size_one_in=0 --acquire_snapshot_one_in=100 --use_multiget=0 --prefixpercent=0 --get_live_files_one_in=1000 --manual_wal_flush_one_in=1000 --pause_background_one_in=1000 --target_file_size_base=524288 --write_buffer_size=524288 --verify_checksum_one_in=1000 --verify_db_one_in=1000
```
- Rehearsal stress test for normal parameter and aggressive parameter to see if such change can find what existing stress test can find (i.e, no regression in testing capability)
- [Ongoing]Try to find new bugs with this change that are not found by current NonBatchedOpsStressTest with no parallel read and write/delete to same key
Reviewed By: ajkr
Differential Revision: D42257258
Pulled By: hx235
fbshipit-source-id: e6fdc18f1fad3753e5ac91731483a644d9b5b6eb
2 years ago
|
|
|
} else {
|
|
|
|
assert(false);
|
|
|
|
}
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
void PrepareTxnDbOptions(SharedState* shared,
|
|
|
|
TransactionDBOptions& txn_db_opts) override {
|
|
|
|
txn_db_opts.rollback_deletion_type_callback =
|
|
|
|
[shared](TransactionDB*, ColumnFamilyHandle*, const Slice& key) {
|
|
|
|
assert(shared);
|
|
|
|
uint64_t key_num = 0;
|
|
|
|
bool ok = GetIntVal(key.ToString(), &key_num);
|
|
|
|
assert(ok);
|
|
|
|
(void)ok;
|
|
|
|
return !shared->AllowsOverwrite(key_num);
|
|
|
|
};
|
|
|
|
}
|
|
|
|
};
|
|
|
|
|
|
|
|
StressTest* CreateNonBatchedOpsStressTest() {
|
|
|
|
return new NonBatchedOpsStressTest();
|
|
|
|
}
|
|
|
|
|
|
|
|
} // namespace ROCKSDB_NAMESPACE
|
|
|
|
#endif // GFLAGS
|