Reduce lock contention for MergeTree tables (by renaming parts without holding lock) by azat · Pull Request #61973 · ClickHouse/ClickHouse

azat · 2024-03-27T11:39:01Z

Changelog category (leave one):

Not for changelog (changelog entry is not required)

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

This change was reverted

Reduce lock contention for MergeTree tables (by renaming parts without holding lock)

Under heavy load, or not so heavy but with fsync_part_directory=1, time that renameTo() holds DataPartsLock will be increased, and this will affect almost every operation with this table.

On one of production clusters I saw ~60 seconds with fsync_part_directory=1.

Move the renameTo() out from the critical section.

robot-ch-test-poll4 · 2024-03-27T11:39:58Z

This is an automated comment for commit 6cfd5b2 with description of existing statuses. It's updated for the latest CI running

❌ Click here to open a full report in a separate page

Check name	Description	Status
A Sync	If it fails, ask a maintainer for help	⏳ pending
CI running	A meta-check that indicates the running CI. Normally, it's in success or pending state. The failed status indicates some problems with the PR	❌ failure
Integration tests	The integration tests report. In parenthesis the package type is given, and in square brackets are the optional part/total tests	❌ failure
Mergeable Check	Checks if all other necessary checks are successful	⏳ pending
Performance Comparison	Measure changes in query performance. The performance test report is described in detail here. In square brackets are the optional part/total tests	❌ failure
Stateless tests	Runs stateless functional tests for ClickHouse binaries built in various configurations -- release, debug, with sanitizers, etc	❌ failure
Upgrade check	Runs stress tests on server version from last release and then tries to upgrade it to the version from the PR. It checks if the new server can successfully startup without any errors, crashes or sanitizer asserts	❌ failure

Successful checks

Check name	Description	Status
AST fuzzer	Runs randomly generated queries to catch program errors. The build type is optionally given in parenthesis. If it fails, ask a maintainer for help	✅ success
ClickBench	Runs [ClickBench](https://github.com/ClickHouse/ClickBench/) with instant-attach table	✅ success
ClickHouse build check	Builds ClickHouse in various configurations for use in further steps. You have to fix the builds that fail. Build logs often has enough information to fix the error, but you might have to reproduce the failure locally. The cmake options can be found in the build log, grepping for cmake. Use these options and follow the general build process	✅ success
Compatibility check	Checks that clickhouse binary runs on distributions with old libc versions. If it fails, ask a maintainer for help	✅ success
Docker keeper image	The check to build and optionally push the mentioned image to docker hub	✅ success
Docker server image	The check to build and optionally push the mentioned image to docker hub	✅ success
Docs check	Builds and tests the documentation	✅ success
Fast test	Normally this is the first check that is ran for a PR. It builds ClickHouse and runs most of stateless functional tests, omitting some. If it fails, further checks are not started until it is fixed. Look at the report to see which tests fail, then reproduce the failure locally as described here	✅ success
Flaky tests	Checks if new added or modified tests are flaky by running them repeatedly, in parallel, with more randomization. Functional tests are run 100 times with address sanitizer, and additional randomization of thread scheduling. Integration tests are run up to 10 times. If at least once a new test has failed, or was too long, this check will be red. We don't allow flaky tests, read the doc	✅ success
Install packages	Checks that the built packages are installable in a clear environment	✅ success
PR Check	Checks correctness of the PR's body	✅ success
Stateful tests	Runs stateful functional tests for ClickHouse binaries built in various configurations -- release, debug, with sanitizers, etc	✅ success
Stress test	Runs stateless functional tests concurrently from several clients to detect concurrency-related errors	✅ success
Style check	Runs a set of checks to keep the code style clean. If some of tests failed, see the related log from the report	✅ success
Unit tests	Runs the unit tests for different release types	✅ success

azat · 2024-03-28T17:18:22Z

Likely related test failures:

Stateless tests (asan) [2/4] — Tests are not finished, fail: 1, passed: 383, skipped: 6
Stateless tests (tsan) [5/5] — Tests are not finished, fail: 3, passed: 302, skipped: 2
Stateless tests (tsan, s3 storage) [5/5] — Tests are not finished, fail: 2, passed: 298, skipped: 4

2024.03.28 06:23:18.782173 [ 937 ] {} <Fatal> : Logical error: 'Trying to remove detached part detached/attaching_all_21_21_0 with path store/b9e/b9ea0283-b7ac-48cd-ab64-54e326a59f96/ in remove function. It shouldn't happen'.

Unrelated test failures:

Stress test (asan) — Found signal in gdb.log

Function X expects N argument to have Y type but receives Nullable(Y) after running IfChainToMultiIf pass. #62062

Stateless tests (asan) [1/4] — fail: 1, passed: 1579, skipped: 17
Stateless tests (debug, s3 storage) [5/6] — fail: 1, passed: 1047, skipped: 27
Stateless tests (msan) [5/6] — fail: 1, passed: 1066, skipped: 8

Flaky test 01287_max_execution_speed with analyzer #56052

Stateless tests (coverage) [2/6] — Some queries hung, fail: 1, passed: 1021, skipped: 3

02443_detach_attach_partition is flaky #54748

Stateless tests (release) — fail: 1, passed: 6280, skipped: 9

02115_rewrite_local_join_right_distribute_table is flaky (under analyzer) #62063

azat · 2024-03-29T16:52:07Z

Stateless tests (coverage) [4/6] — fail: 1, passed: 1092, skipped: 2

00993_system_parts_race_condition_drop_zookeeper

2024-03-29 19:57:24 [d922bb02cd61] 2024.03.29 00:56:54.586084 [ 57841 ] {bf240267-0620-4294-afc1-479c58e6be89} <Error> executeQuery: std::exception. Code: 1001, type: std::__1::__fs::filesystem::filesystem_error, e.what() = filesystem error: in file_size: No such file or directory ["/var/lib/clickhouse/store/ea7/ea7a3fd2-cf47-4ec7-91a5-51c69fba1b95/-8_0_138_2_2/data.cmrk3"]

Unrelated:

Stateless tests (tsan) [3/5] — Tests are not finished, fail: 2, passed: 483, skipped: 6
Stateless tests (msan) [3/6] — Tests are not finished, fail: 1, passed: 367, skipped: 7
Stateless tests (debug) [3/5] — Tests are not finished, fail: 5, passed: 484, skipped: 7
Stateless tests (asan) [1/4] — Tests are not finished, fail: 2, passed: 569, skipped: 8

#62062

azat · 2024-04-02T20:48:03Z

Still some issues:

Stateless tests (debug) [5/5] — Tests are not finished, fail: 9, passed: 361, skipped: 2

2024.04.03 02:03:25.724784 [ 1120 ] {} <Fatal> : Logical error: 'Trying to remove detached part detached/all_200_200_0 with path store/6f2/6f24250b-14de-45e1-8b44-f0f25710765d/ in remove function. It shouldn't happen'.

azat · 2024-04-11T16:52:13Z

Stateless tests (release, old analyzer, s3, DatabaseReplicated) [4/4] — fail: 1, passed: 1539, skipped: 27

03093_bug37909_query_does_not_finish - fixed in 4f6b6e3

azat · 2024-04-15T07:51:38Z

This one is ready. Can someone take a look please?

I have some doubts about that now the part is added to the data_parts_vector before rename and it may leads to removing detached parts, and triggering LOGICAL_ERROR, I've fixed it in the rollback code by checking if it is detached part, then revert it to temporary state, instead of removing it. But in general renaming under lock, is very bad, since it can hold the lock for way more longer time than it should be, and during this time you can't work with the table at all.

azat · 2024-04-29T05:59:12Z

Can someone take a look at this one?

azat · 2024-05-28T15:14:40Z

Kind ping

alexey-milovidov · 2024-06-04T04:06:07Z

@azat, need to fix all the tests.

azat · 2024-06-05T08:19:20Z

@alexey-milovidov tests failures does not looks related, but it is quite old, so I will do a rebase at least.

azat · 2024-06-05T11:04:52Z

ClickHouse special build check — 12/13 artifact groups are OK

tidy - broken in S3Queue rework ordered mode parallel processing #64349 (comment) - fixed - Fix clang-tidy #64827

azat · 2024-06-05T16:37:19Z

Stateless tests (release) — fail: 1, passed: 6629, skipped: 13

02956_rocksdb_bulk_sink - Handle logs from rocksdb by ClickHouse internal logging (and enable debug rocksdb logging on CI) #64856

Signed-off-by: Azat Khuzhin <[email protected]>

…t holding lock) Under heavy load, or not so heavy but with fsync_part_directory=1, time that renameTo() holds DataPartsLock will be increased, and this will affect almost every operation with this table. On one of production clusters I saw ~60 seconds with fsync_part_directory=1. Move the renameTo() out from the critical section. v2: instead of using DataPartsLock.lock.lock()/unlock() move the renameTo() into MergeTreeData::Transaction::commit() Signed-off-by: Azat Khuzhin <[email protected]>

Signed-off-by: Azat Khuzhin <[email protected]>

Since there is an assertion that does not allows to remove detached parts during cleanup, which sounds good in general, but breaks this new code. Signed-off-by: Azat Khuzhin <[email protected]>

The problem was that with this patch set renameMergedTemporaryPart() is called without temporary_directory_lock holded (in MergeTask), since it is reseted just before calling renameMergedTemporaryPart(), and this can be seen in logs: 2024.03.29 19:56:42.126919 [ 1341 ] {ea7a3fd2-cf47-4ec7-91a5-51c69fba1b95::-8_0_138_2_2} <Trace> test_btnct5cr.alter_table_0 (ea7a3fd2-cf47-4ec7-91a5-51c69fba1b95) (MergerMutator): Merged 50 parts: [-8_0_0_0_2, -8_138_138_0] -> -8_0_138_2_2 2024.03.29 19:56:42.127034 [ 1341 ] {ea7a3fd2-cf47-4ec7-91a5-51c69fba1b95::-8_0_138_2_2} <Debug> test_btnct5cr.alter_table_0 (ea7a3fd2-cf47-4ec7-91a5-51c69fba1b95): Committing part -8_0_138_2_2 to zookeeper 2024.03.29 19:56:42.128462 [ 884 ] {} <Warning> test_btnct5cr.alter_table_0 (ea7a3fd2-cf47-4ec7-91a5-51c69fba1b95): Removing temporary directory /var/lib/clickhouse/store/ea7/ea7a3fd2-cf47-4ec7-91a5-51c69fba1b95/tmp_merge_-8_0_138_2_2/ 2024.03.29 19:56:42.128647 [ 1341 ] {ea7a3fd2-cf47-4ec7-91a5-51c69fba1b95::-8_0_138_2_2} <Debug> test_btnct5cr.alter_table_0 (ea7a3fd2-cf47-4ec7-91a5-51c69fba1b95): Part -8_0_138_2_2 committed to zookeeper ... 2024.03.29 19:56:54.586084 [ 57841 ] {bf240267-0620-4294-afc1-479c58e6be89} <Error> executeQuery: std::exception. Code: 1001, type: std::__1::__fs::filesystem::filesystem_error, e.what() = filesystem error: in file_size: No such file or directory ["/var/lib/clickhouse/store/ea7/ea7a3fd2-cf47-4ec7-91a5-51c69fba1b95/-8_0_138_2_2/data.cmrk3"] This should fix failures of 00993_system_parts_race_condition_drop_zookeeper in [1]. [1]: https://s3.amazonaws.com/clickhouse-test-reports/61973/f6f826c85dd5b7bb8db16286fd10dcf441a440f7/stateless_tests__coverage__[4_6].html Though now it looks hackish... Signed-off-by: Azat Khuzhin <[email protected]>

Signed-off-by: Azat Khuzhin <[email protected]>

…d_parts_need_rename CI founds [1]: Logical error: 'precommitted_parts.size() >= precommitted_parts_need_rename.size()' [1]: https://s3.amazonaws.com/clickhouse-test-reports/61973/5c1e6a3e956917bdbb7eaa467934e5b75f17a923/stateless_tests__tsan__s3_storage__[5_5].html The problem is that after precommitted_parts cleaned from detached parts it may be less then precommitted_parts_need_rename, so to avoid this, let's just copy it to a new container. Signed-off-by: Azat Khuzhin <[email protected]>

azat · 2024-06-06T05:33:07Z

Stateless tests (debug, s3 storage) [3/6] fail: 1, passed: 1071, skipped: 24

Fix 03002_part_log_rmt_fetch_mutate_error flakiness #64887

Stateless tests (tsan, s3 storage) [1/5] fail: 1, passed: 1329, skipped: 24

Fix possible loss of "Query was cancelled" message in client (fixes 03023_zeros_generate_random_with_limit_progress_bar flakiness) #64888

alexey-milovidov · 2024-06-06T19:15:44Z

@azat, please resubmit.

azat changed the title ~~[RFC] Rename parts without holding DataPartsLock~~ [RFC] Reduce lock contention for MergeTree tables (by renaming parts without holding DataPartsLock) Mar 27, 2024

robot-ch-test-poll4 added the pr-performance Pull request with some performance improvements label Mar 27, 2024

azat force-pushed the mt/rename-without-lock branch from 3c10ef3 to 21e9798 Compare March 27, 2024 11:39

azat changed the title ~~[RFC] Reduce lock contention for MergeTree tables (by renaming parts without holding DataPartsLock)~~ [RFC] Reduce lock contention for MergeTree tables (by renaming parts without holding lock) Mar 27, 2024

azat marked this pull request as draft March 27, 2024 12:51

azat force-pushed the mt/rename-without-lock branch from 21e9798 to 757c78f Compare March 27, 2024 20:16

azat marked this pull request as ready for review March 27, 2024 20:16

azat marked this pull request as draft March 28, 2024 19:34

azat force-pushed the mt/rename-without-lock branch from 8878b54 to adcf7d5 Compare April 2, 2024 15:33

azat force-pushed the mt/rename-without-lock branch 3 times, most recently from 5c1e6a3 to e08e975 Compare April 11, 2024 12:14

azat marked this pull request as ready for review April 11, 2024 14:43

alexey-milovidov changed the title ~~[RFC] Reduce lock contention for MergeTree tables (by renaming parts without holding lock)~~ Reduce lock contention for MergeTree tables (by renaming parts without holding lock) May 9, 2024

CurtizJ self-assigned this May 9, 2024

alexey-milovidov assigned alexey-milovidov and unassigned CurtizJ May 21, 2024

azat force-pushed the mt/rename-without-lock branch from e08e975 to ec2b3a3 Compare June 5, 2024 08:20

azat force-pushed the mt/rename-without-lock branch from ec2b3a3 to ffa2ab7 Compare June 5, 2024 12:47

azat mentioned this pull request Jun 5, 2024

CI: mergeable check redesign #64093

Merged

31 tasks

azat added 9 commits June 5, 2024 19:38

Add reason into "Part {} is broken and need manual correction" message

66a2962

Signed-off-by: Azat Khuzhin <[email protected]>

Remove unused locked_parts from MergeTreeData::Transaction

6c3db34

Signed-off-by: Azat Khuzhin <[email protected]>

Fix replacing parts with empty

ee546fa

Signed-off-by: Azat Khuzhin <[email protected]>

Use renameParts() explicitly to avoid leaving parts in detached

b41d08a

Since there is an assertion that does not allows to remove detached parts during cleanup, which sounds good in general, but breaks this new code. Signed-off-by: Azat Khuzhin <[email protected]>

Require explicit rename of parts in transaction

3675c27

Signed-off-by: Azat Khuzhin <[email protected]>

Do not remove detached parts in Transaction::rollback

6f522c1

Signed-off-by: Azat Khuzhin <[email protected]>

azat force-pushed the mt/rename-without-lock branch from ffa2ab7 to 6cfd5b2 Compare June 5, 2024 17:38

alexey-milovidov merged commit ce244e1 into ClickHouse:master Jun 6, 2024

azat deleted the mt/rename-without-lock branch June 6, 2024 03:57

alesapin mentioned this pull request Jun 6, 2024

Revert "Reduce lock contention for MergeTree tables (by renaming parts without holding lock)" #64899

Merged

robot-clickhouse-ci-1 added the pr-synced-to-cloud The PR is synced to the cloud repo label Jun 6, 2024

azat mentioned this pull request Jun 7, 2024

Lock-free parts rename for ReplicatedMergeTree to avoid INSERT affect SELECT (resubmit) #64955

Merged

Algunenano added the pr-not-for-changelog This PR should not be mentioned in the changelog label Jun 20, 2024

Conversation

azat commented Mar 27, 2024 • edited by Algunenano Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changelog category (leave one):

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

Uh oh!

robot-ch-test-poll4 commented Mar 27, 2024 • edited by robot-clickhouse Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

azat commented Mar 28, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

azat commented Mar 29, 2024

Uh oh!

azat commented Apr 2, 2024

Uh oh!

azat commented Apr 11, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

azat commented Apr 15, 2024

Uh oh!

azat commented Apr 29, 2024

Uh oh!

azat commented May 28, 2024

Uh oh!

alexey-milovidov commented Jun 4, 2024

Uh oh!

azat commented Jun 5, 2024

Uh oh!

azat commented Jun 5, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

azat commented Jun 5, 2024

Uh oh!

azat commented Jun 6, 2024

Uh oh!

alexey-milovidov commented Jun 6, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

azat commented Mar 27, 2024 •

edited by Algunenano

Loading

robot-ch-test-poll4 commented Mar 27, 2024 •

edited by robot-clickhouse

Loading

azat commented Mar 28, 2024 •

edited

Loading

azat commented Apr 11, 2024 •

edited

Loading

azat commented Jun 5, 2024 •

edited

Loading