|
|
|
// Copyright (c) 2021-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).
|
|
|
|
|
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 <atomic>
|
|
|
|
#ifdef GFLAGS
|
|
|
|
|
|
|
|
|
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
|
|
|
#include "db/wide/wide_column_serialization.h"
|
|
|
|
#include "db_stress_tool/db_stress_common.h"
|
|
|
|
#include "db_stress_tool/db_stress_shared_state.h"
|
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"
|
|
|
|
#include "rocksdb/trace_reader_writer.h"
|
|
|
|
#include "rocksdb/trace_record_result.h"
|
|
|
|
|
|
|
|
namespace ROCKSDB_NAMESPACE {
|
|
|
|
ExpectedState::ExpectedState(size_t max_key, size_t num_column_families)
|
|
|
|
: max_key_(max_key),
|
|
|
|
num_column_families_(num_column_families),
|
|
|
|
values_(nullptr) {}
|
|
|
|
|
|
|
|
void ExpectedState::ClearColumnFamily(int cf) {
|
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 del_mask = ExpectedValue::GetDelMask();
|
|
|
|
std::fill(&Value(cf, 0 /* key */), &Value(cf + 1, 0 /* key */), del_mask);
|
|
|
|
}
|
|
|
|
|
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
|
|
|
void ExpectedState::Precommit(int cf, int64_t key, const ExpectedValue& value) {
|
|
|
|
Value(cf, key).store(value.Read());
|
|
|
|
// To prevent low-level instruction reordering that results
|
|
|
|
// in db write happens before setting pending state in expected value
|
|
|
|
std::atomic_thread_fence(std::memory_order_release);
|
|
|
|
}
|
|
|
|
|
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 ExpectedState::PreparePut(int cf, int64_t key) {
|
|
|
|
ExpectedValue expected_value = Load(cf, key);
|
|
|
|
const ExpectedValue orig_expected_value = expected_value;
|
|
|
|
expected_value.Put(true /* pending */);
|
|
|
|
const ExpectedValue pending_expected_value = expected_value;
|
|
|
|
expected_value.Put(false /* pending */);
|
|
|
|
const ExpectedValue final_expected_value = expected_value;
|
|
|
|
Precommit(cf, key, pending_expected_value);
|
|
|
|
return PendingExpectedValue(&Value(cf, key), orig_expected_value,
|
|
|
|
final_expected_value);
|
|
|
|
}
|
|
|
|
|
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 ExpectedState::Get(int cf, int64_t key) { return Load(cf, key); }
|
|
|
|
|
|
|
|
PendingExpectedValue ExpectedState::PrepareDelete(int cf, int64_t key,
|
|
|
|
bool* prepared) {
|
|
|
|
ExpectedValue expected_value = Load(cf, key);
|
|
|
|
const ExpectedValue orig_expected_value = expected_value;
|
|
|
|
bool res = expected_value.Delete(true /* pending */);
|
|
|
|
if (prepared) {
|
|
|
|
*prepared = res;
|
|
|
|
}
|
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 (!res) {
|
|
|
|
return PendingExpectedValue(&Value(cf, key), orig_expected_value,
|
|
|
|
orig_expected_value);
|
|
|
|
}
|
|
|
|
const ExpectedValue pending_expected_value = expected_value;
|
|
|
|
expected_value.Delete(false /* pending */);
|
|
|
|
const ExpectedValue final_expected_value = expected_value;
|
|
|
|
Precommit(cf, key, pending_expected_value);
|
|
|
|
return PendingExpectedValue(&Value(cf, key), orig_expected_value,
|
|
|
|
final_expected_value);
|
|
|
|
}
|
|
|
|
|
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 ExpectedState::PrepareSingleDelete(int cf, int64_t key) {
|
|
|
|
return PrepareDelete(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
|
|
|
std::vector<PendingExpectedValue> ExpectedState::PrepareDeleteRange(
|
|
|
|
int cf, int64_t begin_key, int64_t end_key) {
|
|
|
|
std::vector<PendingExpectedValue> pending_expected_values;
|
|
|
|
for (int64_t key = begin_key; key < end_key; ++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
|
|
|
bool prepared = false;
|
|
|
|
PendingExpectedValue pending_expected_value =
|
|
|
|
PrepareDelete(cf, key, &prepared);
|
|
|
|
if (prepared) {
|
|
|
|
pending_expected_values.push_back(pending_expected_value);
|
|
|
|
}
|
|
|
|
}
|
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
|
|
|
return pending_expected_values;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool ExpectedState::Exists(int cf, int64_t 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
|
|
|
return Load(cf, key).Exists();
|
|
|
|
}
|
|
|
|
|
|
|
|
void ExpectedState::Reset() {
|
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 del_mask = ExpectedValue::GetDelMask();
|
|
|
|
for (size_t i = 0; i < num_column_families_; ++i) {
|
|
|
|
for (size_t j = 0; j < max_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
|
|
|
Value(static_cast<int>(i), j).store(del_mask, std::memory_order_relaxed);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
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
|
|
|
void ExpectedState::SyncPut(int cf, int64_t key, uint32_t value_base) {
|
|
|
|
ExpectedValue expected_value = Load(cf, key);
|
|
|
|
expected_value.SyncPut(value_base);
|
|
|
|
Value(cf, key).store(expected_value.Read());
|
|
|
|
}
|
|
|
|
|
|
|
|
void ExpectedState::SyncPendingPut(int cf, int64_t key) {
|
|
|
|
ExpectedValue expected_value = Load(cf, key);
|
|
|
|
expected_value.SyncPendingPut();
|
|
|
|
Value(cf, key).store(expected_value.Read());
|
|
|
|
}
|
|
|
|
|
|
|
|
void ExpectedState::SyncDelete(int cf, int64_t key) {
|
|
|
|
ExpectedValue expected_value = Load(cf, key);
|
|
|
|
expected_value.SyncDelete();
|
|
|
|
Value(cf, key).store(expected_value.Read());
|
|
|
|
}
|
|
|
|
|
|
|
|
void ExpectedState::SyncDeleteRange(int cf, int64_t begin_key,
|
|
|
|
int64_t end_key) {
|
|
|
|
for (int64_t key = begin_key; key < end_key; ++key) {
|
|
|
|
SyncDelete(cf, key);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
FileExpectedState::FileExpectedState(std::string expected_state_file_path,
|
|
|
|
size_t max_key, size_t num_column_families)
|
|
|
|
: ExpectedState(max_key, num_column_families),
|
|
|
|
expected_state_file_path_(expected_state_file_path) {}
|
|
|
|
|
|
|
|
Status FileExpectedState::Open(bool create) {
|
|
|
|
size_t expected_values_size = GetValuesLen();
|
|
|
|
|
|
|
|
Env* default_env = Env::Default();
|
|
|
|
|
|
|
|
Status status;
|
|
|
|
if (create) {
|
|
|
|
std::unique_ptr<WritableFile> wfile;
|
|
|
|
const EnvOptions soptions;
|
|
|
|
status = default_env->NewWritableFile(expected_state_file_path_, &wfile,
|
|
|
|
soptions);
|
|
|
|
if (status.ok()) {
|
|
|
|
std::string buf(expected_values_size, '\0');
|
|
|
|
status = wfile->Append(buf);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (status.ok()) {
|
|
|
|
status = default_env->NewMemoryMappedFileBuffer(
|
|
|
|
expected_state_file_path_, &expected_state_mmap_buffer_);
|
|
|
|
}
|
|
|
|
if (status.ok()) {
|
|
|
|
assert(expected_state_mmap_buffer_->GetLen() == expected_values_size);
|
|
|
|
values_ = static_cast<std::atomic<uint32_t>*>(
|
|
|
|
expected_state_mmap_buffer_->GetBase());
|
|
|
|
assert(values_ != nullptr);
|
|
|
|
if (create) {
|
|
|
|
Reset();
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
assert(values_ == nullptr);
|
|
|
|
}
|
|
|
|
return status;
|
|
|
|
}
|
|
|
|
|
|
|
|
AnonExpectedState::AnonExpectedState(size_t max_key, size_t num_column_families)
|
|
|
|
: ExpectedState(max_key, num_column_families) {}
|
|
|
|
|
|
|
|
#ifndef NDEBUG
|
|
|
|
Status AnonExpectedState::Open(bool create) {
|
|
|
|
#else
|
|
|
|
Status AnonExpectedState::Open(bool /* create */) {
|
|
|
|
#endif
|
|
|
|
// AnonExpectedState only supports being freshly created.
|
|
|
|
assert(create);
|
|
|
|
values_allocation_.reset(
|
|
|
|
new std::atomic<uint32_t>[GetValuesLen() /
|
|
|
|
sizeof(std::atomic<uint32_t>)]);
|
|
|
|
values_ = &values_allocation_[0];
|
|
|
|
Reset();
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
ExpectedStateManager::ExpectedStateManager(size_t max_key,
|
|
|
|
size_t num_column_families)
|
|
|
|
: max_key_(max_key),
|
|
|
|
num_column_families_(num_column_families),
|
|
|
|
latest_(nullptr) {}
|
|
|
|
|
|
|
|
ExpectedStateManager::~ExpectedStateManager() {}
|
|
|
|
|
|
|
|
const std::string FileExpectedStateManager::kLatestBasename = "LATEST";
|
|
|
|
const std::string FileExpectedStateManager::kStateFilenameSuffix = ".state";
|
|
|
|
const std::string FileExpectedStateManager::kTraceFilenameSuffix = ".trace";
|
|
|
|
const std::string FileExpectedStateManager::kTempFilenamePrefix = ".";
|
|
|
|
const std::string FileExpectedStateManager::kTempFilenameSuffix = ".tmp";
|
|
|
|
|
|
|
|
FileExpectedStateManager::FileExpectedStateManager(
|
|
|
|
size_t max_key, size_t num_column_families,
|
|
|
|
std::string expected_state_dir_path)
|
|
|
|
: ExpectedStateManager(max_key, num_column_families),
|
|
|
|
expected_state_dir_path_(std::move(expected_state_dir_path)) {
|
|
|
|
assert(!expected_state_dir_path_.empty());
|
|
|
|
}
|
|
|
|
|
|
|
|
Status FileExpectedStateManager::Open() {
|
|
|
|
// Before doing anything, sync directory state with ours. That is, determine
|
|
|
|
// `saved_seqno_`, and create any necessary missing files.
|
|
|
|
std::vector<std::string> expected_state_dir_children;
|
|
|
|
Status s = Env::Default()->GetChildren(expected_state_dir_path_,
|
|
|
|
&expected_state_dir_children);
|
|
|
|
bool found_trace = false;
|
|
|
|
if (s.ok()) {
|
|
|
|
for (size_t i = 0; i < expected_state_dir_children.size(); ++i) {
|
|
|
|
const auto& filename = expected_state_dir_children[i];
|
|
|
|
if (filename.size() >= kStateFilenameSuffix.size() &&
|
|
|
|
filename.rfind(kStateFilenameSuffix) ==
|
|
|
|
filename.size() - kStateFilenameSuffix.size() &&
|
|
|
|
filename.rfind(kLatestBasename, 0) == std::string::npos) {
|
|
|
|
SequenceNumber found_seqno = ParseUint64(
|
|
|
|
filename.substr(0, filename.size() - kStateFilenameSuffix.size()));
|
|
|
|
if (saved_seqno_ == kMaxSequenceNumber || found_seqno > saved_seqno_) {
|
|
|
|
saved_seqno_ = found_seqno;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
// Check if crash happened after creating state file but before creating
|
|
|
|
// trace file.
|
|
|
|
if (saved_seqno_ != kMaxSequenceNumber) {
|
|
|
|
std::string saved_seqno_trace_path = GetPathForFilename(
|
|
|
|
std::to_string(saved_seqno_) + kTraceFilenameSuffix);
|
|
|
|
Status exists_status = Env::Default()->FileExists(saved_seqno_trace_path);
|
|
|
|
if (exists_status.ok()) {
|
|
|
|
found_trace = true;
|
|
|
|
} else if (exists_status.IsNotFound()) {
|
|
|
|
found_trace = false;
|
|
|
|
} else {
|
|
|
|
s = exists_status;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (s.ok() && saved_seqno_ != kMaxSequenceNumber && !found_trace) {
|
|
|
|
// Create an empty trace file so later logic does not need to distinguish
|
|
|
|
// missing vs. empty trace file.
|
|
|
|
std::unique_ptr<WritableFile> wfile;
|
|
|
|
const EnvOptions soptions;
|
|
|
|
std::string saved_seqno_trace_path =
|
|
|
|
GetPathForFilename(std::to_string(saved_seqno_) + kTraceFilenameSuffix);
|
|
|
|
s = Env::Default()->NewWritableFile(saved_seqno_trace_path, &wfile,
|
|
|
|
soptions);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (s.ok()) {
|
|
|
|
s = Clean();
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string expected_state_file_path =
|
|
|
|
GetPathForFilename(kLatestBasename + kStateFilenameSuffix);
|
|
|
|
bool found = false;
|
|
|
|
if (s.ok()) {
|
|
|
|
Status exists_status = Env::Default()->FileExists(expected_state_file_path);
|
|
|
|
if (exists_status.ok()) {
|
|
|
|
found = true;
|
|
|
|
} else if (exists_status.IsNotFound()) {
|
|
|
|
found = false;
|
|
|
|
} else {
|
|
|
|
s = exists_status;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!found) {
|
|
|
|
// Initialize the file in a temp path and then rename it. That way, in case
|
|
|
|
// this process is killed during setup, `Clean()` will take care of removing
|
|
|
|
// the incomplete expected values file.
|
|
|
|
std::string temp_expected_state_file_path =
|
|
|
|
GetTempPathForFilename(kLatestBasename + kStateFilenameSuffix);
|
|
|
|
FileExpectedState temp_expected_state(temp_expected_state_file_path,
|
|
|
|
max_key_, num_column_families_);
|
|
|
|
if (s.ok()) {
|
|
|
|
s = temp_expected_state.Open(true /* create */);
|
|
|
|
}
|
|
|
|
if (s.ok()) {
|
|
|
|
s = Env::Default()->RenameFile(temp_expected_state_file_path,
|
|
|
|
expected_state_file_path);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (s.ok()) {
|
|
|
|
latest_.reset(new FileExpectedState(std::move(expected_state_file_path),
|
|
|
|
max_key_, num_column_families_));
|
|
|
|
s = latest_->Open(false /* create */);
|
|
|
|
}
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
Status FileExpectedStateManager::SaveAtAndAfter(DB* db) {
|
|
|
|
SequenceNumber seqno = db->GetLatestSequenceNumber();
|
|
|
|
|
|
|
|
std::string state_filename = std::to_string(seqno) + kStateFilenameSuffix;
|
|
|
|
std::string state_file_temp_path = GetTempPathForFilename(state_filename);
|
|
|
|
std::string state_file_path = GetPathForFilename(state_filename);
|
|
|
|
|
|
|
|
std::string latest_file_path =
|
|
|
|
GetPathForFilename(kLatestBasename + kStateFilenameSuffix);
|
|
|
|
|
|
|
|
std::string trace_filename = std::to_string(seqno) + kTraceFilenameSuffix;
|
|
|
|
std::string trace_file_path = GetPathForFilename(trace_filename);
|
|
|
|
|
|
|
|
// Populate a tempfile and then rename it to atomically create "<seqno>.state"
|
|
|
|
// with contents from "LATEST.state"
|
|
|
|
Status s = CopyFile(FileSystem::Default(), latest_file_path,
|
|
|
|
state_file_temp_path, 0 /* size */, false /* use_fsync */,
|
|
|
|
nullptr /* io_tracer */, Temperature::kUnknown);
|
|
|
|
if (s.ok()) {
|
|
|
|
s = FileSystem::Default()->RenameFile(state_file_temp_path, state_file_path,
|
|
|
|
IOOptions(), nullptr /* dbg */);
|
|
|
|
}
|
|
|
|
SequenceNumber old_saved_seqno = 0;
|
|
|
|
if (s.ok()) {
|
|
|
|
old_saved_seqno = saved_seqno_;
|
|
|
|
saved_seqno_ = seqno;
|
|
|
|
}
|
|
|
|
|
|
|
|
// If there is a crash now, i.e., after "<seqno>.state" was created but before
|
|
|
|
// "<seqno>.trace" is created, it will be treated as if "<seqno>.trace" were
|
|
|
|
// present but empty.
|
|
|
|
|
|
|
|
// Create "<seqno>.trace" directly. It is initially empty so no need for
|
|
|
|
// tempfile.
|
|
|
|
std::unique_ptr<TraceWriter> trace_writer;
|
|
|
|
if (s.ok()) {
|
|
|
|
EnvOptions soptions;
|
|
|
|
// Disable buffering so traces will not get stuck in application buffer.
|
|
|
|
soptions.writable_file_max_buffer_size = 0;
|
|
|
|
s = NewFileTraceWriter(Env::Default(), soptions, trace_file_path,
|
|
|
|
&trace_writer);
|
|
|
|
}
|
|
|
|
if (s.ok()) {
|
|
|
|
TraceOptions trace_opts;
|
|
|
|
trace_opts.filter |= kTraceFilterGet;
|
|
|
|
trace_opts.filter |= kTraceFilterMultiGet;
|
|
|
|
trace_opts.filter |= kTraceFilterIteratorSeek;
|
|
|
|
trace_opts.filter |= kTraceFilterIteratorSeekForPrev;
|
|
|
|
trace_opts.preserve_write_order = true;
|
|
|
|
s = db->StartTrace(trace_opts, std::move(trace_writer));
|
|
|
|
}
|
|
|
|
|
|
|
|
// Delete old state/trace files. Deletion order does not matter since we only
|
|
|
|
// delete after successfully saving new files, so old files will never be used
|
|
|
|
// again, even if we crash.
|
|
|
|
if (s.ok() && old_saved_seqno != kMaxSequenceNumber &&
|
|
|
|
old_saved_seqno != saved_seqno_) {
|
|
|
|
s = Env::Default()->DeleteFile(GetPathForFilename(
|
|
|
|
std::to_string(old_saved_seqno) + kStateFilenameSuffix));
|
|
|
|
}
|
|
|
|
if (s.ok() && old_saved_seqno != kMaxSequenceNumber &&
|
|
|
|
old_saved_seqno != saved_seqno_) {
|
|
|
|
s = Env::Default()->DeleteFile(GetPathForFilename(
|
|
|
|
std::to_string(old_saved_seqno) + kTraceFilenameSuffix));
|
|
|
|
}
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool FileExpectedStateManager::HasHistory() {
|
|
|
|
return saved_seqno_ != kMaxSequenceNumber;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
namespace {
|
|
|
|
|
|
|
|
// An `ExpectedStateTraceRecordHandler` applies a configurable number of
|
|
|
|
// write operation trace records to the configured expected state. It is used in
|
|
|
|
// `FileExpectedStateManager::Restore()` to sync the expected state with the
|
|
|
|
// DB's post-recovery state.
|
|
|
|
class ExpectedStateTraceRecordHandler : public TraceRecord::Handler,
|
|
|
|
public WriteBatch::Handler {
|
|
|
|
public:
|
|
|
|
ExpectedStateTraceRecordHandler(uint64_t max_write_ops, ExpectedState* state)
|
Support WriteCommit policy with sync_fault_injection=1 (#10624)
Summary:
**Context:**
Prior to this PR, correctness testing with un-sync data loss [disabled](https://github.com/facebook/rocksdb/pull/10605) transaction (`use_txn=1`) thus all of the `txn_write_policy` . This PR improved that by adding support for one policy - WriteCommit (`txn_write_policy=0`).
**Summary:**
They key to this support is (a) handle Mark{Begin, End}Prepare/MarkCommit/MarkRollback in constructing ExpectedState under WriteCommit policy correctly and (b) monitor CI jobs and solve any test incompatibility issue till jobs are stable. (b) will be part of the test plan.
For (a)
- During prepare (i.e, between `MarkBeginPrepare()` and `MarkEndPrepare(xid)`), `ExpectedStateTraceRecordHandler` will buffer all writes by adding all writes to an internal `WriteBatch`.
- On `MarkEndPrepare()`, that `WriteBatch` will be associated with the transaction's `xid`.
- During the commit (i.e, on `MarkCommit(xid)`), `ExpectedStateTraceRecordHandler` will retrieve and iterate the internal `WriteBatch` and finally apply those writes to `ExpectedState`
- During the rollback (i.e, on `MarkRollback(xid)`), `ExpectedStateTraceRecordHandler` will erase the internal `WriteBatch` from the map.
For (b) - one major issue described below:
- TransactionsDB in db stress recovers prepared-but-not-committed txns from the previous crashed run by randomly committing or rolling back it at the start of the current run, see a historical [PR](https://github.com/facebook/rocksdb/commit/6d06be22c083ccf185fd38dba49fde73b644b4c1) predated correctness testing.
- And we will verify those processed keys in a recovered db against their expected state.
- However since now we turn on `sync_fault_injection=1` where the expected state is constructed from the trace instead of using the LATEST.state from previous run. The expected state now used to verify those processed keys won't contain UNKNOWN_SENTINEL as they should - see test 1 for a failed case.
- Therefore, we decided to manually update its expected state to be UNKNOWN_SENTINEL as part of the processing.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10624
Test Plan:
1. Test exposed the major issue described above. This test will fail without setting UNKNOWN_SENTINEL in expected state during the processing and pass after
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
dbt=$db.tmp
expt=$exp.tmp
rm -rf $db $exp
mkdir -p $exp
echo "RUN 1"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 2"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 3"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
2. Manual testing to ensure ExpectedState is constructed correctly during recovery by verifying it against previously crashed TransactionDB's WAL.
- Run the following command to crash a TransactionDB with WriteCommit policy. Then `./ldb dump_wal` on its WAL file
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
rm -rf $db $exp
mkdir -p $exp
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 30
kill $pid
sleep 1
```
- Run the following command to verify recovery of the crashed db under debugger. Compare the step-wise result with WAL records (e.g, WriteBatch content, xid, prepare/commit/rollback marker)
```
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
3. Automatic testing by triggering all RocksDB stress/crash test jobs for 3 rounds with no failure.
Reviewed By: ajkr, riversand963
Differential Revision: D39199373
Pulled By: hx235
fbshipit-source-id: 7a1dec0e3e2ee6ea86ddf5dd19ceb5543a3d6f0c
2 years ago
|
|
|
: max_write_ops_(max_write_ops),
|
|
|
|
state_(state),
|
|
|
|
buffered_writes_(nullptr) {}
|
|
|
|
|
|
|
|
~ExpectedStateTraceRecordHandler() { assert(IsDone()); }
|
|
|
|
|
|
|
|
// True if we have already reached the limit on write operations to apply.
|
|
|
|
bool IsDone() { return num_write_ops_ == max_write_ops_; }
|
|
|
|
|
|
|
|
Status Handle(const WriteQueryTraceRecord& record,
|
|
|
|
std::unique_ptr<TraceRecordResult>* /* result */) override {
|
|
|
|
if (IsDone()) {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
WriteBatch batch(record.GetWriteBatchRep().ToString());
|
|
|
|
return batch.Iterate(this);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Ignore reads.
|
|
|
|
Status Handle(const GetQueryTraceRecord& /* record */,
|
|
|
|
std::unique_ptr<TraceRecordResult>* /* result */) override {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
// Ignore reads.
|
|
|
|
Status Handle(const IteratorSeekQueryTraceRecord& /* record */,
|
|
|
|
std::unique_ptr<TraceRecordResult>* /* result */) override {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
// Ignore reads.
|
|
|
|
Status Handle(const MultiGetQueryTraceRecord& /* record */,
|
|
|
|
std::unique_ptr<TraceRecordResult>* /* result */) override {
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
// Below are the WriteBatch::Handler overrides. We could use a separate
|
|
|
|
// object, but it's convenient and works to share state with the
|
|
|
|
// `TraceRecord::Handler`.
|
|
|
|
|
|
|
|
Status PutCF(uint32_t column_family_id, const Slice& key_with_ts,
|
|
|
|
const Slice& value) override {
|
|
|
|
Slice key =
|
|
|
|
StripTimestampFromUserKey(key_with_ts, FLAGS_user_timestamp_size);
|
|
|
|
uint64_t key_id;
|
|
|
|
if (!GetIntVal(key.ToString(), &key_id)) {
|
|
|
|
return Status::Corruption("unable to parse key", key.ToString());
|
|
|
|
}
|
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(value);
|
|
|
|
|
Support WriteCommit policy with sync_fault_injection=1 (#10624)
Summary:
**Context:**
Prior to this PR, correctness testing with un-sync data loss [disabled](https://github.com/facebook/rocksdb/pull/10605) transaction (`use_txn=1`) thus all of the `txn_write_policy` . This PR improved that by adding support for one policy - WriteCommit (`txn_write_policy=0`).
**Summary:**
They key to this support is (a) handle Mark{Begin, End}Prepare/MarkCommit/MarkRollback in constructing ExpectedState under WriteCommit policy correctly and (b) monitor CI jobs and solve any test incompatibility issue till jobs are stable. (b) will be part of the test plan.
For (a)
- During prepare (i.e, between `MarkBeginPrepare()` and `MarkEndPrepare(xid)`), `ExpectedStateTraceRecordHandler` will buffer all writes by adding all writes to an internal `WriteBatch`.
- On `MarkEndPrepare()`, that `WriteBatch` will be associated with the transaction's `xid`.
- During the commit (i.e, on `MarkCommit(xid)`), `ExpectedStateTraceRecordHandler` will retrieve and iterate the internal `WriteBatch` and finally apply those writes to `ExpectedState`
- During the rollback (i.e, on `MarkRollback(xid)`), `ExpectedStateTraceRecordHandler` will erase the internal `WriteBatch` from the map.
For (b) - one major issue described below:
- TransactionsDB in db stress recovers prepared-but-not-committed txns from the previous crashed run by randomly committing or rolling back it at the start of the current run, see a historical [PR](https://github.com/facebook/rocksdb/commit/6d06be22c083ccf185fd38dba49fde73b644b4c1) predated correctness testing.
- And we will verify those processed keys in a recovered db against their expected state.
- However since now we turn on `sync_fault_injection=1` where the expected state is constructed from the trace instead of using the LATEST.state from previous run. The expected state now used to verify those processed keys won't contain UNKNOWN_SENTINEL as they should - see test 1 for a failed case.
- Therefore, we decided to manually update its expected state to be UNKNOWN_SENTINEL as part of the processing.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10624
Test Plan:
1. Test exposed the major issue described above. This test will fail without setting UNKNOWN_SENTINEL in expected state during the processing and pass after
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
dbt=$db.tmp
expt=$exp.tmp
rm -rf $db $exp
mkdir -p $exp
echo "RUN 1"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 2"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 3"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
2. Manual testing to ensure ExpectedState is constructed correctly during recovery by verifying it against previously crashed TransactionDB's WAL.
- Run the following command to crash a TransactionDB with WriteCommit policy. Then `./ldb dump_wal` on its WAL file
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
rm -rf $db $exp
mkdir -p $exp
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 30
kill $pid
sleep 1
```
- Run the following command to verify recovery of the crashed db under debugger. Compare the step-wise result with WAL records (e.g, WriteBatch content, xid, prepare/commit/rollback marker)
```
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
3. Automatic testing by triggering all RocksDB stress/crash test jobs for 3 rounds with no failure.
Reviewed By: ajkr, riversand963
Differential Revision: D39199373
Pulled By: hx235
fbshipit-source-id: 7a1dec0e3e2ee6ea86ddf5dd19ceb5543a3d6f0c
2 years ago
|
|
|
bool should_buffer_write = !(buffered_writes_ == nullptr);
|
|
|
|
if (should_buffer_write) {
|
|
|
|
return WriteBatchInternal::Put(buffered_writes_.get(), column_family_id,
|
|
|
|
key, value);
|
|
|
|
}
|
|
|
|
|
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
|
|
|
state_->SyncPut(column_family_id, static_cast<int64_t>(key_id), value_base);
|
|
|
|
++num_write_ops_;
|
|
|
|
return Status::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
|
|
|
Status PutEntityCF(uint32_t column_family_id, const Slice& key_with_ts,
|
|
|
|
const Slice& entity) override {
|
|
|
|
Slice key =
|
|
|
|
StripTimestampFromUserKey(key_with_ts, FLAGS_user_timestamp_size);
|
|
|
|
|
|
|
|
uint64_t key_id = 0;
|
|
|
|
if (!GetIntVal(key.ToString(), &key_id)) {
|
|
|
|
return Status::Corruption("Unable to parse key", key.ToString());
|
|
|
|
}
|
|
|
|
|
|
|
|
Slice entity_copy = entity;
|
|
|
|
WideColumns columns;
|
|
|
|
if (!WideColumnSerialization::Deserialize(entity_copy, columns).ok()) {
|
|
|
|
return Status::Corruption("Unable to deserialize entity",
|
|
|
|
entity.ToString(/* hex */ true));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!VerifyWideColumns(columns)) {
|
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
|
|
|
return Status::Corruption("Wide columns in entity inconsistent",
|
|
|
|
entity.ToString(/* hex */ true));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (buffered_writes_) {
|
|
|
|
return WriteBatchInternal::PutEntity(buffered_writes_.get(),
|
|
|
|
column_family_id, key, columns);
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(!columns.empty());
|
|
|
|
assert(columns.front().name() == kDefaultWideColumnName);
|
|
|
|
|
|
|
|
const uint32_t value_base = GetValueBase(columns.front().value());
|
|
|
|
|
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
|
|
|
state_->SyncPut(column_family_id, static_cast<int64_t>(key_id), value_base);
|
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
|
|
|
|
|
|
|
++num_write_ops_;
|
|
|
|
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
Status DeleteCF(uint32_t column_family_id,
|
|
|
|
const Slice& key_with_ts) override {
|
|
|
|
Slice key =
|
|
|
|
StripTimestampFromUserKey(key_with_ts, FLAGS_user_timestamp_size);
|
|
|
|
uint64_t key_id;
|
|
|
|
if (!GetIntVal(key.ToString(), &key_id)) {
|
|
|
|
return Status::Corruption("unable to parse key", key.ToString());
|
|
|
|
}
|
|
|
|
|
Support WriteCommit policy with sync_fault_injection=1 (#10624)
Summary:
**Context:**
Prior to this PR, correctness testing with un-sync data loss [disabled](https://github.com/facebook/rocksdb/pull/10605) transaction (`use_txn=1`) thus all of the `txn_write_policy` . This PR improved that by adding support for one policy - WriteCommit (`txn_write_policy=0`).
**Summary:**
They key to this support is (a) handle Mark{Begin, End}Prepare/MarkCommit/MarkRollback in constructing ExpectedState under WriteCommit policy correctly and (b) monitor CI jobs and solve any test incompatibility issue till jobs are stable. (b) will be part of the test plan.
For (a)
- During prepare (i.e, between `MarkBeginPrepare()` and `MarkEndPrepare(xid)`), `ExpectedStateTraceRecordHandler` will buffer all writes by adding all writes to an internal `WriteBatch`.
- On `MarkEndPrepare()`, that `WriteBatch` will be associated with the transaction's `xid`.
- During the commit (i.e, on `MarkCommit(xid)`), `ExpectedStateTraceRecordHandler` will retrieve and iterate the internal `WriteBatch` and finally apply those writes to `ExpectedState`
- During the rollback (i.e, on `MarkRollback(xid)`), `ExpectedStateTraceRecordHandler` will erase the internal `WriteBatch` from the map.
For (b) - one major issue described below:
- TransactionsDB in db stress recovers prepared-but-not-committed txns from the previous crashed run by randomly committing or rolling back it at the start of the current run, see a historical [PR](https://github.com/facebook/rocksdb/commit/6d06be22c083ccf185fd38dba49fde73b644b4c1) predated correctness testing.
- And we will verify those processed keys in a recovered db against their expected state.
- However since now we turn on `sync_fault_injection=1` where the expected state is constructed from the trace instead of using the LATEST.state from previous run. The expected state now used to verify those processed keys won't contain UNKNOWN_SENTINEL as they should - see test 1 for a failed case.
- Therefore, we decided to manually update its expected state to be UNKNOWN_SENTINEL as part of the processing.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10624
Test Plan:
1. Test exposed the major issue described above. This test will fail without setting UNKNOWN_SENTINEL in expected state during the processing and pass after
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
dbt=$db.tmp
expt=$exp.tmp
rm -rf $db $exp
mkdir -p $exp
echo "RUN 1"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 2"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 3"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
2. Manual testing to ensure ExpectedState is constructed correctly during recovery by verifying it against previously crashed TransactionDB's WAL.
- Run the following command to crash a TransactionDB with WriteCommit policy. Then `./ldb dump_wal` on its WAL file
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
rm -rf $db $exp
mkdir -p $exp
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 30
kill $pid
sleep 1
```
- Run the following command to verify recovery of the crashed db under debugger. Compare the step-wise result with WAL records (e.g, WriteBatch content, xid, prepare/commit/rollback marker)
```
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
3. Automatic testing by triggering all RocksDB stress/crash test jobs for 3 rounds with no failure.
Reviewed By: ajkr, riversand963
Differential Revision: D39199373
Pulled By: hx235
fbshipit-source-id: 7a1dec0e3e2ee6ea86ddf5dd19ceb5543a3d6f0c
2 years ago
|
|
|
bool should_buffer_write = !(buffered_writes_ == nullptr);
|
|
|
|
if (should_buffer_write) {
|
|
|
|
return WriteBatchInternal::Delete(buffered_writes_.get(),
|
|
|
|
column_family_id, 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
|
|
|
state_->SyncDelete(column_family_id, static_cast<int64_t>(key_id));
|
|
|
|
++num_write_ops_;
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
Status SingleDeleteCF(uint32_t column_family_id,
|
|
|
|
const Slice& key_with_ts) override {
|
Support WriteCommit policy with sync_fault_injection=1 (#10624)
Summary:
**Context:**
Prior to this PR, correctness testing with un-sync data loss [disabled](https://github.com/facebook/rocksdb/pull/10605) transaction (`use_txn=1`) thus all of the `txn_write_policy` . This PR improved that by adding support for one policy - WriteCommit (`txn_write_policy=0`).
**Summary:**
They key to this support is (a) handle Mark{Begin, End}Prepare/MarkCommit/MarkRollback in constructing ExpectedState under WriteCommit policy correctly and (b) monitor CI jobs and solve any test incompatibility issue till jobs are stable. (b) will be part of the test plan.
For (a)
- During prepare (i.e, between `MarkBeginPrepare()` and `MarkEndPrepare(xid)`), `ExpectedStateTraceRecordHandler` will buffer all writes by adding all writes to an internal `WriteBatch`.
- On `MarkEndPrepare()`, that `WriteBatch` will be associated with the transaction's `xid`.
- During the commit (i.e, on `MarkCommit(xid)`), `ExpectedStateTraceRecordHandler` will retrieve and iterate the internal `WriteBatch` and finally apply those writes to `ExpectedState`
- During the rollback (i.e, on `MarkRollback(xid)`), `ExpectedStateTraceRecordHandler` will erase the internal `WriteBatch` from the map.
For (b) - one major issue described below:
- TransactionsDB in db stress recovers prepared-but-not-committed txns from the previous crashed run by randomly committing or rolling back it at the start of the current run, see a historical [PR](https://github.com/facebook/rocksdb/commit/6d06be22c083ccf185fd38dba49fde73b644b4c1) predated correctness testing.
- And we will verify those processed keys in a recovered db against their expected state.
- However since now we turn on `sync_fault_injection=1` where the expected state is constructed from the trace instead of using the LATEST.state from previous run. The expected state now used to verify those processed keys won't contain UNKNOWN_SENTINEL as they should - see test 1 for a failed case.
- Therefore, we decided to manually update its expected state to be UNKNOWN_SENTINEL as part of the processing.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10624
Test Plan:
1. Test exposed the major issue described above. This test will fail without setting UNKNOWN_SENTINEL in expected state during the processing and pass after
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
dbt=$db.tmp
expt=$exp.tmp
rm -rf $db $exp
mkdir -p $exp
echo "RUN 1"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 2"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 3"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
2. Manual testing to ensure ExpectedState is constructed correctly during recovery by verifying it against previously crashed TransactionDB's WAL.
- Run the following command to crash a TransactionDB with WriteCommit policy. Then `./ldb dump_wal` on its WAL file
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
rm -rf $db $exp
mkdir -p $exp
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 30
kill $pid
sleep 1
```
- Run the following command to verify recovery of the crashed db under debugger. Compare the step-wise result with WAL records (e.g, WriteBatch content, xid, prepare/commit/rollback marker)
```
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
3. Automatic testing by triggering all RocksDB stress/crash test jobs for 3 rounds with no failure.
Reviewed By: ajkr, riversand963
Differential Revision: D39199373
Pulled By: hx235
fbshipit-source-id: 7a1dec0e3e2ee6ea86ddf5dd19ceb5543a3d6f0c
2 years ago
|
|
|
bool should_buffer_write = !(buffered_writes_ == nullptr);
|
|
|
|
if (should_buffer_write) {
|
|
|
|
Slice key =
|
|
|
|
StripTimestampFromUserKey(key_with_ts, FLAGS_user_timestamp_size);
|
|
|
|
Slice ts =
|
|
|
|
ExtractTimestampFromUserKey(key_with_ts, FLAGS_user_timestamp_size);
|
|
|
|
std::array<Slice, 2> key_with_ts_arr{{key, ts}};
|
|
|
|
return WriteBatchInternal::SingleDelete(
|
|
|
|
buffered_writes_.get(), column_family_id,
|
|
|
|
SliceParts(key_with_ts_arr.data(), 2));
|
|
|
|
}
|
|
|
|
|
|
|
|
return DeleteCF(column_family_id, key_with_ts);
|
|
|
|
}
|
|
|
|
|
|
|
|
Status DeleteRangeCF(uint32_t column_family_id,
|
|
|
|
const Slice& begin_key_with_ts,
|
|
|
|
const Slice& end_key_with_ts) override {
|
|
|
|
Slice begin_key =
|
|
|
|
StripTimestampFromUserKey(begin_key_with_ts, FLAGS_user_timestamp_size);
|
|
|
|
Slice end_key =
|
|
|
|
StripTimestampFromUserKey(end_key_with_ts, FLAGS_user_timestamp_size);
|
|
|
|
uint64_t begin_key_id, end_key_id;
|
|
|
|
if (!GetIntVal(begin_key.ToString(), &begin_key_id)) {
|
|
|
|
return Status::Corruption("unable to parse begin key",
|
|
|
|
begin_key.ToString());
|
|
|
|
}
|
|
|
|
if (!GetIntVal(end_key.ToString(), &end_key_id)) {
|
|
|
|
return Status::Corruption("unable to parse end key", end_key.ToString());
|
|
|
|
}
|
|
|
|
|
Support WriteCommit policy with sync_fault_injection=1 (#10624)
Summary:
**Context:**
Prior to this PR, correctness testing with un-sync data loss [disabled](https://github.com/facebook/rocksdb/pull/10605) transaction (`use_txn=1`) thus all of the `txn_write_policy` . This PR improved that by adding support for one policy - WriteCommit (`txn_write_policy=0`).
**Summary:**
They key to this support is (a) handle Mark{Begin, End}Prepare/MarkCommit/MarkRollback in constructing ExpectedState under WriteCommit policy correctly and (b) monitor CI jobs and solve any test incompatibility issue till jobs are stable. (b) will be part of the test plan.
For (a)
- During prepare (i.e, between `MarkBeginPrepare()` and `MarkEndPrepare(xid)`), `ExpectedStateTraceRecordHandler` will buffer all writes by adding all writes to an internal `WriteBatch`.
- On `MarkEndPrepare()`, that `WriteBatch` will be associated with the transaction's `xid`.
- During the commit (i.e, on `MarkCommit(xid)`), `ExpectedStateTraceRecordHandler` will retrieve and iterate the internal `WriteBatch` and finally apply those writes to `ExpectedState`
- During the rollback (i.e, on `MarkRollback(xid)`), `ExpectedStateTraceRecordHandler` will erase the internal `WriteBatch` from the map.
For (b) - one major issue described below:
- TransactionsDB in db stress recovers prepared-but-not-committed txns from the previous crashed run by randomly committing or rolling back it at the start of the current run, see a historical [PR](https://github.com/facebook/rocksdb/commit/6d06be22c083ccf185fd38dba49fde73b644b4c1) predated correctness testing.
- And we will verify those processed keys in a recovered db against their expected state.
- However since now we turn on `sync_fault_injection=1` where the expected state is constructed from the trace instead of using the LATEST.state from previous run. The expected state now used to verify those processed keys won't contain UNKNOWN_SENTINEL as they should - see test 1 for a failed case.
- Therefore, we decided to manually update its expected state to be UNKNOWN_SENTINEL as part of the processing.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10624
Test Plan:
1. Test exposed the major issue described above. This test will fail without setting UNKNOWN_SENTINEL in expected state during the processing and pass after
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
dbt=$db.tmp
expt=$exp.tmp
rm -rf $db $exp
mkdir -p $exp
echo "RUN 1"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 2"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 3"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
2. Manual testing to ensure ExpectedState is constructed correctly during recovery by verifying it against previously crashed TransactionDB's WAL.
- Run the following command to crash a TransactionDB with WriteCommit policy. Then `./ldb dump_wal` on its WAL file
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
rm -rf $db $exp
mkdir -p $exp
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 30
kill $pid
sleep 1
```
- Run the following command to verify recovery of the crashed db under debugger. Compare the step-wise result with WAL records (e.g, WriteBatch content, xid, prepare/commit/rollback marker)
```
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
3. Automatic testing by triggering all RocksDB stress/crash test jobs for 3 rounds with no failure.
Reviewed By: ajkr, riversand963
Differential Revision: D39199373
Pulled By: hx235
fbshipit-source-id: 7a1dec0e3e2ee6ea86ddf5dd19ceb5543a3d6f0c
2 years ago
|
|
|
bool should_buffer_write = !(buffered_writes_ == nullptr);
|
|
|
|
if (should_buffer_write) {
|
|
|
|
return WriteBatchInternal::DeleteRange(
|
|
|
|
buffered_writes_.get(), column_family_id, begin_key, end_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
|
|
|
state_->SyncDeleteRange(column_family_id,
|
|
|
|
static_cast<int64_t>(begin_key_id),
|
|
|
|
static_cast<int64_t>(end_key_id));
|
|
|
|
++num_write_ops_;
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
Status MergeCF(uint32_t column_family_id, const Slice& key_with_ts,
|
|
|
|
const Slice& value) override {
|
|
|
|
Slice key =
|
|
|
|
StripTimestampFromUserKey(key_with_ts, FLAGS_user_timestamp_size);
|
Support WriteCommit policy with sync_fault_injection=1 (#10624)
Summary:
**Context:**
Prior to this PR, correctness testing with un-sync data loss [disabled](https://github.com/facebook/rocksdb/pull/10605) transaction (`use_txn=1`) thus all of the `txn_write_policy` . This PR improved that by adding support for one policy - WriteCommit (`txn_write_policy=0`).
**Summary:**
They key to this support is (a) handle Mark{Begin, End}Prepare/MarkCommit/MarkRollback in constructing ExpectedState under WriteCommit policy correctly and (b) monitor CI jobs and solve any test incompatibility issue till jobs are stable. (b) will be part of the test plan.
For (a)
- During prepare (i.e, between `MarkBeginPrepare()` and `MarkEndPrepare(xid)`), `ExpectedStateTraceRecordHandler` will buffer all writes by adding all writes to an internal `WriteBatch`.
- On `MarkEndPrepare()`, that `WriteBatch` will be associated with the transaction's `xid`.
- During the commit (i.e, on `MarkCommit(xid)`), `ExpectedStateTraceRecordHandler` will retrieve and iterate the internal `WriteBatch` and finally apply those writes to `ExpectedState`
- During the rollback (i.e, on `MarkRollback(xid)`), `ExpectedStateTraceRecordHandler` will erase the internal `WriteBatch` from the map.
For (b) - one major issue described below:
- TransactionsDB in db stress recovers prepared-but-not-committed txns from the previous crashed run by randomly committing or rolling back it at the start of the current run, see a historical [PR](https://github.com/facebook/rocksdb/commit/6d06be22c083ccf185fd38dba49fde73b644b4c1) predated correctness testing.
- And we will verify those processed keys in a recovered db against their expected state.
- However since now we turn on `sync_fault_injection=1` where the expected state is constructed from the trace instead of using the LATEST.state from previous run. The expected state now used to verify those processed keys won't contain UNKNOWN_SENTINEL as they should - see test 1 for a failed case.
- Therefore, we decided to manually update its expected state to be UNKNOWN_SENTINEL as part of the processing.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10624
Test Plan:
1. Test exposed the major issue described above. This test will fail without setting UNKNOWN_SENTINEL in expected state during the processing and pass after
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
dbt=$db.tmp
expt=$exp.tmp
rm -rf $db $exp
mkdir -p $exp
echo "RUN 1"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 2"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 3"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
2. Manual testing to ensure ExpectedState is constructed correctly during recovery by verifying it against previously crashed TransactionDB's WAL.
- Run the following command to crash a TransactionDB with WriteCommit policy. Then `./ldb dump_wal` on its WAL file
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
rm -rf $db $exp
mkdir -p $exp
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 30
kill $pid
sleep 1
```
- Run the following command to verify recovery of the crashed db under debugger. Compare the step-wise result with WAL records (e.g, WriteBatch content, xid, prepare/commit/rollback marker)
```
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
3. Automatic testing by triggering all RocksDB stress/crash test jobs for 3 rounds with no failure.
Reviewed By: ajkr, riversand963
Differential Revision: D39199373
Pulled By: hx235
fbshipit-source-id: 7a1dec0e3e2ee6ea86ddf5dd19ceb5543a3d6f0c
2 years ago
|
|
|
|
|
|
|
bool should_buffer_write = !(buffered_writes_ == nullptr);
|
|
|
|
if (should_buffer_write) {
|
|
|
|
return WriteBatchInternal::Merge(buffered_writes_.get(), column_family_id,
|
|
|
|
key, value);
|
|
|
|
}
|
|
|
|
|
|
|
|
return PutCF(column_family_id, key, value);
|
|
|
|
}
|
|
|
|
|
Support WriteCommit policy with sync_fault_injection=1 (#10624)
Summary:
**Context:**
Prior to this PR, correctness testing with un-sync data loss [disabled](https://github.com/facebook/rocksdb/pull/10605) transaction (`use_txn=1`) thus all of the `txn_write_policy` . This PR improved that by adding support for one policy - WriteCommit (`txn_write_policy=0`).
**Summary:**
They key to this support is (a) handle Mark{Begin, End}Prepare/MarkCommit/MarkRollback in constructing ExpectedState under WriteCommit policy correctly and (b) monitor CI jobs and solve any test incompatibility issue till jobs are stable. (b) will be part of the test plan.
For (a)
- During prepare (i.e, between `MarkBeginPrepare()` and `MarkEndPrepare(xid)`), `ExpectedStateTraceRecordHandler` will buffer all writes by adding all writes to an internal `WriteBatch`.
- On `MarkEndPrepare()`, that `WriteBatch` will be associated with the transaction's `xid`.
- During the commit (i.e, on `MarkCommit(xid)`), `ExpectedStateTraceRecordHandler` will retrieve and iterate the internal `WriteBatch` and finally apply those writes to `ExpectedState`
- During the rollback (i.e, on `MarkRollback(xid)`), `ExpectedStateTraceRecordHandler` will erase the internal `WriteBatch` from the map.
For (b) - one major issue described below:
- TransactionsDB in db stress recovers prepared-but-not-committed txns from the previous crashed run by randomly committing or rolling back it at the start of the current run, see a historical [PR](https://github.com/facebook/rocksdb/commit/6d06be22c083ccf185fd38dba49fde73b644b4c1) predated correctness testing.
- And we will verify those processed keys in a recovered db against their expected state.
- However since now we turn on `sync_fault_injection=1` where the expected state is constructed from the trace instead of using the LATEST.state from previous run. The expected state now used to verify those processed keys won't contain UNKNOWN_SENTINEL as they should - see test 1 for a failed case.
- Therefore, we decided to manually update its expected state to be UNKNOWN_SENTINEL as part of the processing.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10624
Test Plan:
1. Test exposed the major issue described above. This test will fail without setting UNKNOWN_SENTINEL in expected state during the processing and pass after
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
dbt=$db.tmp
expt=$exp.tmp
rm -rf $db $exp
mkdir -p $exp
echo "RUN 1"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 2"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 3"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
2. Manual testing to ensure ExpectedState is constructed correctly during recovery by verifying it against previously crashed TransactionDB's WAL.
- Run the following command to crash a TransactionDB with WriteCommit policy. Then `./ldb dump_wal` on its WAL file
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
rm -rf $db $exp
mkdir -p $exp
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 30
kill $pid
sleep 1
```
- Run the following command to verify recovery of the crashed db under debugger. Compare the step-wise result with WAL records (e.g, WriteBatch content, xid, prepare/commit/rollback marker)
```
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
3. Automatic testing by triggering all RocksDB stress/crash test jobs for 3 rounds with no failure.
Reviewed By: ajkr, riversand963
Differential Revision: D39199373
Pulled By: hx235
fbshipit-source-id: 7a1dec0e3e2ee6ea86ddf5dd19ceb5543a3d6f0c
2 years ago
|
|
|
Status MarkBeginPrepare(bool = false) override {
|
|
|
|
assert(!buffered_writes_);
|
|
|
|
buffered_writes_.reset(new WriteBatch());
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
Status MarkEndPrepare(const Slice& xid) override {
|
|
|
|
assert(buffered_writes_);
|
|
|
|
std::string xid_str = xid.ToString();
|
|
|
|
assert(xid_to_buffered_writes_.find(xid_str) ==
|
|
|
|
xid_to_buffered_writes_.end());
|
|
|
|
|
|
|
|
xid_to_buffered_writes_[xid_str].swap(buffered_writes_);
|
|
|
|
|
|
|
|
buffered_writes_.reset();
|
|
|
|
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
Status MarkCommit(const Slice& xid) override {
|
|
|
|
std::string xid_str = xid.ToString();
|
|
|
|
assert(xid_to_buffered_writes_.find(xid_str) !=
|
|
|
|
xid_to_buffered_writes_.end());
|
|
|
|
assert(xid_to_buffered_writes_.at(xid_str));
|
|
|
|
|
|
|
|
Status s = xid_to_buffered_writes_.at(xid_str)->Iterate(this);
|
|
|
|
xid_to_buffered_writes_.erase(xid_str);
|
|
|
|
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
Status MarkRollback(const Slice& xid) override {
|
|
|
|
std::string xid_str = xid.ToString();
|
|
|
|
assert(xid_to_buffered_writes_.find(xid_str) !=
|
|
|
|
xid_to_buffered_writes_.end());
|
|
|
|
assert(xid_to_buffered_writes_.at(xid_str));
|
|
|
|
xid_to_buffered_writes_.erase(xid_str);
|
|
|
|
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
|
|
|
|
private:
|
|
|
|
uint64_t num_write_ops_ = 0;
|
|
|
|
uint64_t max_write_ops_;
|
|
|
|
ExpectedState* state_;
|
Support WriteCommit policy with sync_fault_injection=1 (#10624)
Summary:
**Context:**
Prior to this PR, correctness testing with un-sync data loss [disabled](https://github.com/facebook/rocksdb/pull/10605) transaction (`use_txn=1`) thus all of the `txn_write_policy` . This PR improved that by adding support for one policy - WriteCommit (`txn_write_policy=0`).
**Summary:**
They key to this support is (a) handle Mark{Begin, End}Prepare/MarkCommit/MarkRollback in constructing ExpectedState under WriteCommit policy correctly and (b) monitor CI jobs and solve any test incompatibility issue till jobs are stable. (b) will be part of the test plan.
For (a)
- During prepare (i.e, between `MarkBeginPrepare()` and `MarkEndPrepare(xid)`), `ExpectedStateTraceRecordHandler` will buffer all writes by adding all writes to an internal `WriteBatch`.
- On `MarkEndPrepare()`, that `WriteBatch` will be associated with the transaction's `xid`.
- During the commit (i.e, on `MarkCommit(xid)`), `ExpectedStateTraceRecordHandler` will retrieve and iterate the internal `WriteBatch` and finally apply those writes to `ExpectedState`
- During the rollback (i.e, on `MarkRollback(xid)`), `ExpectedStateTraceRecordHandler` will erase the internal `WriteBatch` from the map.
For (b) - one major issue described below:
- TransactionsDB in db stress recovers prepared-but-not-committed txns from the previous crashed run by randomly committing or rolling back it at the start of the current run, see a historical [PR](https://github.com/facebook/rocksdb/commit/6d06be22c083ccf185fd38dba49fde73b644b4c1) predated correctness testing.
- And we will verify those processed keys in a recovered db against their expected state.
- However since now we turn on `sync_fault_injection=1` where the expected state is constructed from the trace instead of using the LATEST.state from previous run. The expected state now used to verify those processed keys won't contain UNKNOWN_SENTINEL as they should - see test 1 for a failed case.
- Therefore, we decided to manually update its expected state to be UNKNOWN_SENTINEL as part of the processing.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10624
Test Plan:
1. Test exposed the major issue described above. This test will fail without setting UNKNOWN_SENTINEL in expected state during the processing and pass after
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
dbt=$db.tmp
expt=$exp.tmp
rm -rf $db $exp
mkdir -p $exp
echo "RUN 1"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 2"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 0.2
sleep 20
kill $pid
sleep 0.2
echo "RUN 3"
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
2. Manual testing to ensure ExpectedState is constructed correctly during recovery by verifying it against previously crashed TransactionDB's WAL.
- Run the following command to crash a TransactionDB with WriteCommit policy. Then `./ldb dump_wal` on its WAL file
```
db=/dev/shm/rocksdb_crashtest_blackbox
exp=/dev/shm/rocksdb_crashtest_expected
rm -rf $db $exp
mkdir -p $exp
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1 &
pid=$!
sleep 30
kill $pid
sleep 1
```
- Run the following command to verify recovery of the crashed db under debugger. Compare the step-wise result with WAL records (e.g, WriteBatch content, xid, prepare/commit/rollback marker)
```
./db_stress \
--clear_column_family_one_in=0 --column_families=1 --db=$db --delpercent=10 --delrangepercent=0 --destroy_db_initially=0 --expected_values_dir=$exp --iterpercent=0 --key_len_percent_dist=1,30,69 --max_key=1000000 --max_key_len=3 --prefixpercent=0 --readpercent=0 --reopen=0 --ops_per_thread=100000000 --test_batches_snapshots=0 --value_size_mult=32 --writepercent=90 \
--use_txn=1 --txn_write_policy=0 --sync_fault_injection=1
```
3. Automatic testing by triggering all RocksDB stress/crash test jobs for 3 rounds with no failure.
Reviewed By: ajkr, riversand963
Differential Revision: D39199373
Pulled By: hx235
fbshipit-source-id: 7a1dec0e3e2ee6ea86ddf5dd19ceb5543a3d6f0c
2 years ago
|
|
|
std::unordered_map<std::string, std::unique_ptr<WriteBatch>>
|
|
|
|
xid_to_buffered_writes_;
|
|
|
|
std::unique_ptr<WriteBatch> buffered_writes_;
|
|
|
|
};
|
|
|
|
|
|
|
|
} // anonymous namespace
|
|
|
|
|
|
|
|
Status FileExpectedStateManager::Restore(DB* db) {
|
|
|
|
assert(HasHistory());
|
|
|
|
SequenceNumber seqno = db->GetLatestSequenceNumber();
|
|
|
|
if (seqno < saved_seqno_) {
|
|
|
|
return Status::Corruption("DB is older than any restorable expected state");
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string state_filename =
|
|
|
|
std::to_string(saved_seqno_) + kStateFilenameSuffix;
|
|
|
|
std::string state_file_path = GetPathForFilename(state_filename);
|
|
|
|
|
|
|
|
std::string latest_file_temp_path =
|
|
|
|
GetTempPathForFilename(kLatestBasename + kStateFilenameSuffix);
|
|
|
|
std::string latest_file_path =
|
|
|
|
GetPathForFilename(kLatestBasename + kStateFilenameSuffix);
|
|
|
|
|
|
|
|
std::string trace_filename =
|
|
|
|
std::to_string(saved_seqno_) + kTraceFilenameSuffix;
|
|
|
|
std::string trace_file_path = GetPathForFilename(trace_filename);
|
|
|
|
|
|
|
|
std::unique_ptr<TraceReader> trace_reader;
|
|
|
|
Status s = NewFileTraceReader(Env::Default(), EnvOptions(), trace_file_path,
|
|
|
|
&trace_reader);
|
|
|
|
|
|
|
|
if (s.ok()) {
|
|
|
|
// We are going to replay on top of "`seqno`.state" to create a new
|
|
|
|
// "LATEST.state". Start off by creating a tempfile so we can later make the
|
|
|
|
// new "LATEST.state" appear atomically using `RenameFile()`.
|
|
|
|
s = CopyFile(FileSystem::Default(), state_file_path, latest_file_temp_path,
|
|
|
|
0 /* size */, false /* use_fsync */, nullptr /* io_tracer */,
|
|
|
|
Temperature::kUnknown);
|
|
|
|
}
|
|
|
|
|
|
|
|
{
|
|
|
|
std::unique_ptr<Replayer> replayer;
|
|
|
|
std::unique_ptr<ExpectedState> state;
|
|
|
|
std::unique_ptr<ExpectedStateTraceRecordHandler> handler;
|
|
|
|
if (s.ok()) {
|
|
|
|
state.reset(new FileExpectedState(latest_file_temp_path, max_key_,
|
|
|
|
num_column_families_));
|
|
|
|
s = state->Open(false /* create */);
|
|
|
|
}
|
|
|
|
if (s.ok()) {
|
|
|
|
handler.reset(new ExpectedStateTraceRecordHandler(seqno - saved_seqno_,
|
|
|
|
state.get()));
|
|
|
|
// TODO(ajkr): An API limitation requires we provide `handles` although
|
|
|
|
// they will be unused since we only use the replayer for reading records.
|
|
|
|
// Just give a default CFH for now to satisfy the requirement.
|
|
|
|
s = db->NewDefaultReplayer({db->DefaultColumnFamily()} /* handles */,
|
|
|
|
std::move(trace_reader), &replayer);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (s.ok()) {
|
|
|
|
s = replayer->Prepare();
|
|
|
|
}
|
|
|
|
for (;;) {
|
|
|
|
std::unique_ptr<TraceRecord> record;
|
|
|
|
s = replayer->Next(&record);
|
|
|
|
if (!s.ok()) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
std::unique_ptr<TraceRecordResult> res;
|
|
|
|
record->Accept(handler.get(), &res);
|
|
|
|
}
|
|
|
|
if (s.IsCorruption() && handler->IsDone()) {
|
|
|
|
// There could be a corruption reading the tail record of the trace due to
|
|
|
|
// `db_stress` crashing while writing it. It shouldn't matter as long as
|
|
|
|
// we already found all the write ops we need to catch up the expected
|
|
|
|
// state.
|
|
|
|
s = Status::OK();
|
|
|
|
}
|
|
|
|
if (s.IsIncomplete()) {
|
|
|
|
// OK because `Status::Incomplete` is expected upon finishing all the
|
|
|
|
// trace records.
|
|
|
|
s = Status::OK();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (s.ok()) {
|
|
|
|
s = FileSystem::Default()->RenameFile(latest_file_temp_path,
|
|
|
|
latest_file_path, IOOptions(),
|
|
|
|
nullptr /* dbg */);
|
|
|
|
}
|
|
|
|
if (s.ok()) {
|
|
|
|
latest_.reset(new FileExpectedState(latest_file_path, max_key_,
|
|
|
|
num_column_families_));
|
|
|
|
s = latest_->Open(false /* create */);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Delete old state/trace files. We must delete the state file first.
|
|
|
|
// Otherwise, a crash-recovery immediately after deleting the trace file could
|
|
|
|
// lead to `Restore()` unable to replay to `seqno`.
|
|
|
|
if (s.ok()) {
|
|
|
|
s = Env::Default()->DeleteFile(state_file_path);
|
|
|
|
}
|
|
|
|
if (s.ok()) {
|
|
|
|
saved_seqno_ = kMaxSequenceNumber;
|
|
|
|
s = Env::Default()->DeleteFile(trace_file_path);
|
|
|
|
}
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
Status FileExpectedStateManager::Clean() {
|
|
|
|
std::vector<std::string> expected_state_dir_children;
|
|
|
|
Status s = Env::Default()->GetChildren(expected_state_dir_path_,
|
|
|
|
&expected_state_dir_children);
|
|
|
|
// An incomplete `Open()` or incomplete `SaveAtAndAfter()` could have left
|
|
|
|
// behind invalid temporary files. An incomplete `SaveAtAndAfter()` could have
|
|
|
|
// also left behind stale state/trace files. An incomplete `Restore()` could
|
|
|
|
// have left behind stale trace files.
|
|
|
|
for (size_t i = 0; s.ok() && i < expected_state_dir_children.size(); ++i) {
|
|
|
|
const auto& filename = expected_state_dir_children[i];
|
|
|
|
if (filename.rfind(kTempFilenamePrefix, 0 /* pos */) == 0 &&
|
|
|
|
filename.size() >= kTempFilenameSuffix.size() &&
|
|
|
|
filename.rfind(kTempFilenameSuffix) ==
|
|
|
|
filename.size() - kTempFilenameSuffix.size()) {
|
|
|
|
// Delete all temp files.
|
|
|
|
s = Env::Default()->DeleteFile(GetPathForFilename(filename));
|
|
|
|
} else if (filename.size() >= kStateFilenameSuffix.size() &&
|
|
|
|
filename.rfind(kStateFilenameSuffix) ==
|
|
|
|
filename.size() - kStateFilenameSuffix.size() &&
|
|
|
|
filename.rfind(kLatestBasename, 0) == std::string::npos &&
|
|
|
|
ParseUint64(filename.substr(
|
|
|
|
0, filename.size() - kStateFilenameSuffix.size())) <
|
|
|
|
saved_seqno_) {
|
|
|
|
assert(saved_seqno_ != kMaxSequenceNumber);
|
|
|
|
// Delete stale state files.
|
|
|
|
s = Env::Default()->DeleteFile(GetPathForFilename(filename));
|
|
|
|
} else if (filename.size() >= kTraceFilenameSuffix.size() &&
|
|
|
|
filename.rfind(kTraceFilenameSuffix) ==
|
|
|
|
filename.size() - kTraceFilenameSuffix.size() &&
|
|
|
|
ParseUint64(filename.substr(
|
|
|
|
0, filename.size() - kTraceFilenameSuffix.size())) <
|
|
|
|
saved_seqno_) {
|
|
|
|
// Delete stale trace files.
|
|
|
|
s = Env::Default()->DeleteFile(GetPathForFilename(filename));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string FileExpectedStateManager::GetTempPathForFilename(
|
|
|
|
const std::string& filename) {
|
|
|
|
assert(!expected_state_dir_path_.empty());
|
|
|
|
std::string expected_state_dir_path_slash =
|
|
|
|
expected_state_dir_path_.back() == '/' ? expected_state_dir_path_
|
|
|
|
: expected_state_dir_path_ + "/";
|
|
|
|
return expected_state_dir_path_slash + kTempFilenamePrefix + filename +
|
|
|
|
kTempFilenameSuffix;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string FileExpectedStateManager::GetPathForFilename(
|
|
|
|
const std::string& filename) {
|
|
|
|
assert(!expected_state_dir_path_.empty());
|
|
|
|
std::string expected_state_dir_path_slash =
|
|
|
|
expected_state_dir_path_.back() == '/' ? expected_state_dir_path_
|
|
|
|
: expected_state_dir_path_ + "/";
|
|
|
|
return expected_state_dir_path_slash + filename;
|
|
|
|
}
|
|
|
|
|
|
|
|
AnonExpectedStateManager::AnonExpectedStateManager(size_t max_key,
|
|
|
|
size_t num_column_families)
|
|
|
|
: ExpectedStateManager(max_key, num_column_families) {}
|
|
|
|
|
|
|
|
Status AnonExpectedStateManager::Open() {
|
|
|
|
latest_.reset(new AnonExpectedState(max_key_, num_column_families_));
|
|
|
|
return latest_->Open(true /* create */);
|
|
|
|
}
|
|
|
|
|
|
|
|
} // namespace ROCKSDB_NAMESPACE
|
|
|
|
|
|
|
|
#endif // GFLAGS
|