Fix RWLock inconsistency after write lock timeout #57454

vitlibar · 2023-12-03T13:09:05Z

Changelog category:

Bug Fix

Changelog entry:

Fix RWLock inconsistency after write lock timeout.

This PR is to provide a correct fix for the issue described in the description of #38864. The fix provided in #38864 was not complete and after that fix after exclusive lock failure RWLock could become inconsistent and not be actually unlocked after releasing all locks. See #42719 (comment)

robot-ch-test-poll · 2023-12-03T13:11:29Z

This is an automated comment for commit 179a0a2 with description of existing statuses. It's updated for the latest CI running

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

Successful checks

Check name	Description	Status
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 image for servers	The check to build and optionally push the mentioned image to docker hub	✅ 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. Integrational 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
Mergeable Check	Checks if all other necessary checks are successful	✅ success
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	✅ success
Push to Dockerhub	The check for building and pushing the CI related docker images to docker hub	✅ success
SQLTest	There's no description for the check yet, please add it to tests/ci/ci_config.py:CHECK_DESCRIPTIONS	✅ success
SQLancer	Fuzzing tests that detect logical bugs with SQLancer tool	✅ success
Sqllogic	Run clickhouse on the sqllogic test set against sqlite and checks that all statements are passed	✅ 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

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	❌ failure
Bugfix validate check	Checks that either a new test (functional or integration) or there some changed tests that fail with the binary built on master branch	❌ error
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	⏳ pending
Integration tests	The integration tests report. In parenthesis the package type is given, and 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

vitlibar · 2023-12-04T13:03:15Z

@alesapin It says Bugfix validate check — Changed tests don't reproduce the bug, but that's not true because the bug is reproduced well in unit-tests (src/unit_tests_dbms). So I suppose we should fix CI to make it consider those unit tests results too, do you know how to do that?

Algunenano

In general it looks great. I've left some comments about minor things

Algunenano · 2023-12-05T10:57:08Z

src/Common/RWLock.cpp

+}
+
+
+String RWLockImpl::getOwnerQueryIdsDescription() const


I'd recommend fmt::formatter to do things like this. It makes formatting containers much simpler.

Not necessary for the PR, just something to keep in mind

Here we build our output in a cycle, fmt::formatter() doesn't seem to be much better in this case.

Algunenano · 2023-12-05T11:15:41Z

tests/integration/test_backup_restore_on_cluster/test_concurrency.py

            )

    def truncate_tables():
        while time.time() < end_time:
            table_name = f"mydb.tbl{randint(1, num_nodes)}"
            node = nodes[randint(0, num_nodes - 1)]
-            node.query(f"TRUNCATE TABLE IF EXISTS {table_name} SYNC")
+            # "TRUNCATE TABLE IF EXISTS" still can throw some errors (e.g. "WRITE locking attempt on node0 has timed out!")


I guess this depends on the table type right? MergeTree should not require an exclusive lock for truncate as it's going to replace the old parts with an empty range.

OTOH In the case of DROP it depends on the database type (Atomic is fine)

For MergeTree-based tables TRUNCATE doesn't require an exclusive lock (see) however this test can create "Log" tables too. I added a comment about that to the test.

Algunenano · 2023-12-05T11:18:51Z

src/Storages/IStorage.cpp

@@ -41,8 +41,8 @@ RWLockImpl::LockHolder IStorage::tryLockTimed(
    {
        const String type_str = type == RWLockImpl::Type::Read ? "READ" : "WRITE";
        throw Exception(ErrorCodes::DEADLOCK_AVOIDED,
-            "{} locking attempt on \"{}\" has timed out! ({}ms) Possible deadlock avoided. Client should retry",
-            type_str, getStorageID(), acquire_timeout.count());
+            "{} locking attempt on \"{}\" has timed out! ({}ms) Possible deadlock avoided. Client should retry. Owner query ids: {}",


Although this might be inconsistent (time to check vs time to read) it's an amazing QOL improvement. It'd have been extremely useful to have it in the past.

Algunenano · 2023-12-05T13:09:42Z

src/Common/RWLock.cpp

@@ -169,11 +178,12 @@ RWLockImpl::getLock(RWLockImpl::Type type, const String & query_id, const std::c
    if (rdlock_owner == readers_queue.end() && wrlock_owner == writers_queue.end())
    {
        (type == Read ? rdlock_owner : wrlock_owner) = it_group;  /// SM2: nothrow


This ternary conditional seems wrong, as this can only be reached if type == Read. Otherwise writers_queue won't be empty

readers_queue and writers_queue can't be both empty here, at least they must contain it_group.
But we don't check those queues for emptyness here, we assign rdlock_owner or wrlock_owner to it_group.
This condition is not wrong.

Algunenano · 2023-12-05T13:28:55Z

src/Common/RWLock.cpp

+{
+    if (rdlock_owner != readers_queue.end())
+    {
+        for (;;)


There should be, at most, only one group of readers pending right?

You can have N readers that are already owner, a writers, and then only a block of readers all together. AFAICS it's not possible to have more than one group of readers that aren't owners, but I might be missing something

Yes, we append readers to the same reader group if this group doesn't have ownership yet.
We create another reading group only if the last reading group is already an owner.
But there can be multiple reader groups with ownership.

I've added chassert() to check that.

Algunenano · 2023-12-05T13:31:19Z

src/Common/RWLock.cpp

            wrlock_owner = writers_queue.begin();
-        }
        else


Is this even possible anymore? If we removed all readers there must be a writer or nothing in the queue, but never another reader group.

There can be an active reader group, then a writer, then another (inactive) reading group.
We remove the active reader group, then we activate the writer.

Algunenano · 2023-12-05T13:51:04Z

src/Common/tests/gtest_rw_lock.cpp

+            if (timepoint.length() < 5)
+                timepoint.insert(0, 5 - timepoint.length(), ' ');
+            std::lock_guard lock{mutex};
+            std::cout << timepoint << " : " << event << std::endl;


We can remove the prints now and leave the asserts

vitlibar · 2023-12-10T12:35:26Z

Test failures:

alexey-milovidov · 2023-12-10T20:52:47Z

@vitlibar a test has failed: https://s3.amazonaws.com/clickhouse-test-reports/57712/6b55c16b4ed40864aa0577fa61a9a6a41c12912d/unit_tests__msan_.html

I've temporarily reverted it.

robot-ch-test-poll added the pr-bugfix Pull request with bugfix, not backported by default label Dec 3, 2023

vitlibar force-pushed the fix-rwlock branch from 7d0eb8c to 7d859a0 Compare December 3, 2023 15:00

vitlibar mentioned this pull request Dec 3, 2023

Flaky test test_backup_restore_on_cluster/test_concurrency.py #42719

Closed

vitlibar force-pushed the fix-rwlock branch 3 times, most recently from 6c21a70 to d5c043d Compare December 3, 2023 20:44

vitlibar added 4 commits December 4, 2023 00:15

Add more tests for RWLock.

fefbf39

Fix RWLock inconsistency after write lock timeout.

44d6605

Show owner query ids in the message for the DEADLOCK_AVOIDED error.

0457aba

Fix test test_create_or_drop_tables_during_backup.

366095a

vitlibar force-pushed the fix-rwlock branch from d5c043d to 366095a Compare December 3, 2023 23:15

Algunenano self-assigned this Dec 5, 2023

Algunenano approved these changes Dec 5, 2023

View reviewed changes

vitlibar force-pushed the fix-rwlock branch 3 times, most recently from 49415fe to ec5348a Compare December 7, 2023 21:07

Add some comments.

b4cac92

vitlibar force-pushed the fix-rwlock branch from ec5348a to b4cac92 Compare December 7, 2023 21:19

Fixed condition and add more chasserts for consistency checks.

179a0a2

vitlibar force-pushed the fix-rwlock branch from c7b0ff8 to 179a0a2 Compare December 9, 2023 19:24

vitlibar merged commit a058a26 into ClickHouse:master Dec 10, 2023
326 of 337 checks passed

vitlibar deleted the fix-rwlock branch December 10, 2023 13:09

alexey-milovidov mentioned this pull request Dec 10, 2023

Revert "Fix RWLock inconsistency after write lock timeout" #57730

Merged

vitlibar restored the fix-rwlock branch December 10, 2023 23:48

vitlibar mentioned this pull request Dec 10, 2023

Fix RWLock inconsistency after write lock timeout (again) #57733

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix RWLock inconsistency after write lock timeout #57454

Fix RWLock inconsistency after write lock timeout #57454

vitlibar commented Dec 3, 2023 •

edited

robot-ch-test-poll commented Dec 3, 2023 •

edited by robot-clickhouse-ci-2

vitlibar commented Dec 4, 2023

Algunenano left a comment

Algunenano Dec 5, 2023

vitlibar Dec 7, 2023 •

edited

Algunenano Dec 5, 2023

Algunenano Dec 5, 2023

vitlibar Dec 7, 2023 •

edited

Algunenano Dec 5, 2023

Algunenano Dec 5, 2023

vitlibar Dec 7, 2023

Algunenano Dec 5, 2023

vitlibar Dec 7, 2023 •

edited

vitlibar Dec 7, 2023

Algunenano Dec 5, 2023

vitlibar Dec 7, 2023

Algunenano Dec 5, 2023

vitlibar Dec 7, 2023

vitlibar commented Dec 10, 2023 •

edited

alexey-milovidov commented Dec 10, 2023

Fix RWLock inconsistency after write lock timeout #57454

Fix RWLock inconsistency after write lock timeout #57454

Conversation

vitlibar commented Dec 3, 2023 • edited

Changelog category:

Changelog entry:

robot-ch-test-poll commented Dec 3, 2023 • edited by robot-clickhouse-ci-2

vitlibar commented Dec 4, 2023

Algunenano left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

vitlibar Dec 7, 2023 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

vitlibar Dec 7, 2023 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

vitlibar Dec 7, 2023 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

vitlibar commented Dec 10, 2023 • edited

alexey-milovidov commented Dec 10, 2023

vitlibar commented Dec 3, 2023 •

edited

robot-ch-test-poll commented Dec 3, 2023 •

edited by robot-clickhouse-ci-2

vitlibar Dec 7, 2023 •

edited

vitlibar Dec 7, 2023 •

edited

vitlibar Dec 7, 2023 •

edited

vitlibar commented Dec 10, 2023 •

edited