kvcoord: deflake BenchmarkTxnWriteBuffer #160648
Merged
+0
−2
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The `SendLocked` subtest has been consistently flaky since the test was introduced. The suspicion is that the ratio of time spent between stop/start is extremely small compared to outside the timed section. This tends to cause the benchmark not to be able to reach the 1s target, as 1s of the timed part could equal 20 minutes in total outside of the timed section. This commit removes the starting and stopping of the timer. The downside is that the benchmark now also measures the allocations as part of `makeBuffer`. But it also makes this subtest consistent with the second subtest (`flushBufferAndSendBatch`) in that respect. Fixes: cockroachdb#151650 Release note: None
Member
|
This change is |
stevendanna
approved these changes
Jan 9, 2026
Contributor
Author
|
bors r=stevendanna |
craig bot
pushed a commit
that referenced
this pull request
Jan 9, 2026
158639: storepool: consider StoreLiveness status when determining suspect stores r=miraradeva a=arulajmani First 4 commits from #158629; only the last one is relevant for this PR. Feel free to hold off on reviewing until the thing merges. ---- This patch treats stores from which we recently (within the suspect duration) withdrew support in StoreLiveness as suspect. Doing so ensures we don't transfer leases to such stores, as shown by the changes to TestLeaseTransferAfterStoreLivenessSupportWithdrawn. Closes #158513 Release note: None 160642: authors: add amerani to authors r=amerani a=amerani Epic: None Release note: None 160647: kvnemesis: disable SQL ops in safety mode r=stevendanna a=miraradeva Currently, there is a single kvnemesis operation that executes via SQL: `ToggleGlobalReads`. We have seen this operation get stuck and cause the test to timeout under safety more. The expected behavior is that any stuck operations would time out, but it seems like there is a context cancelation propagation issue, most likely in lib/pq, but not confirmed. This commit disables `ToggleGlobalReads` in safety mode to reduce test failures. This change would also help confirm that this is the only operation susceptible to the hanging behavior. Fixes: #160293 Release note: None 160648: kvcoord: deflake BenchmarkTxnWriteBuffer r=stevendanna a=miraradeva The `SendLocked` subtest has been consistently flaky since the test was introduced. The suspicion is that the ratio of time spent between stop/start is extremely small compared to outside the timed section. This tends to cause the benchmark not to be able to reach the 1s target, as 1s of the timed part could equal 20 minutes in total outside of the timed section. This commit removes the starting and stopping of the timer. The downside is that the benchmark now also measures the allocations as part of `makeBuffer`. But it also makes this subtest consistent with the second subtest (`flushBufferAndSendBatch`) in that respect. Fixes: #151650 Release note: None 160778: authors: add eric.alton to authors r=eric-alton a=eric-alton Epic: None Release note: None Co-authored-by: Arul Ajmani <[email protected]> Co-authored-by: Alek Merani <[email protected]> Co-authored-by: Mira Radeva <[email protected]> Co-authored-by: eric-alton <[email protected]>
Contributor
|
Build failed (retrying...): |
craig bot
pushed a commit
that referenced
this pull request
Jan 9, 2026
158639: storepool: consider StoreLiveness status when determining suspect stores r=miraradeva a=arulajmani First 4 commits from #158629; only the last one is relevant for this PR. Feel free to hold off on reviewing until the thing merges. ---- This patch treats stores from which we recently (within the suspect duration) withdrew support in StoreLiveness as suspect. Doing so ensures we don't transfer leases to such stores, as shown by the changes to TestLeaseTransferAfterStoreLivenessSupportWithdrawn. Closes #158513 Release note: None 160648: kvcoord: deflake BenchmarkTxnWriteBuffer r=stevendanna a=miraradeva The `SendLocked` subtest has been consistently flaky since the test was introduced. The suspicion is that the ratio of time spent between stop/start is extremely small compared to outside the timed section. This tends to cause the benchmark not to be able to reach the 1s target, as 1s of the timed part could equal 20 minutes in total outside of the timed section. This commit removes the starting and stopping of the timer. The downside is that the benchmark now also measures the allocations as part of `makeBuffer`. But it also makes this subtest consistent with the second subtest (`flushBufferAndSendBatch`) in that respect. Fixes: #151650 Release note: None 160746: scripts/gceworker: add -4 to recent dig patch r=yuzefovich a=yuzefovich Epic: None Release note: None 160778: authors: add eric.alton to authors r=eric-alton a=eric-alton Epic: None Release note: None 160783: storage: refactor TestMVCCHistories r=annrpom a=jbowens Refactor some of TestMVCCHistories to pull more of the logic of executing a run command out of the main Test func. This is in preparation for adding a variant that injects I/O errors into the iterator operations and ensures an error is bubbled up to the caller. Epic: PEBBLE-1158 Informs #139746. Release note: None Co-authored-by: Arul Ajmani <[email protected]> Co-authored-by: Mira Radeva <[email protected]> Co-authored-by: Yahor Yuzefovich <[email protected]> Co-authored-by: eric-alton <[email protected]> Co-authored-by: Jackson Owens <[email protected]>
Contributor
|
Build failed (retrying...):
|
Contributor
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The
SendLockedsubtest has been consistently flaky since the test was introduced. The suspicion is that the ratio of time spent between stop/start is extremely small compared to outside the timed section. This tends to cause the benchmark not to be able to reach the 1s target, as 1s of the timed part could equal 20 minutes in total outside of the timed section.This commit removes the starting and stopping of the timer. The downside is that the benchmark now also measures the allocations as part of
makeBuffer. But it also makes this subtest consistent with the second subtest (flushBufferAndSendBatch) in that respect.Fixes: #151650
Release note: None