Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

backupccl: set shorter closed_timestamp settings in datadriven tests #78489

Merged
merged 1 commit into from
Mar 27, 2022

Conversation

adityamaru
Copy link
Contributor

This change drops the kv.closed_timestamp.side_transport_interval
and kv.closed_timestamp.target_duration in the servers started as part
of the datadriven framework. This change was motivated by the fact that
cluster settings set in certain multi-node datadriven tests were not being
observed by all nodes in the cluster. Cluster settings are now backed by
rangefeeds instead of gossiping the SystemSpanConfig and so these settings
should speed up when all nodes become aware of changes to their values.

Fixes: #77264

Release note: None

@adityamaru adityamaru requested review from stevendanna and a team March 25, 2022 09:28
@adityamaru
Copy link
Contributor Author

adityamaru commented Mar 25, 2022

The symptom was very similar to #77221. There is a more heavy handed solution here where we reject datadriven tests from running SET CLUSTER SETTING sql queries, and instead introduce a new instruction that is backed by serverutils.SetClusterSetting introduced by @miretskiy in the PR linked above. For the time being though I've held off on that patch, since this seems to fix it for us too. Will be worth tracking #73399 to see if something more sophisticated is built out in the future.

Stressed for 15mins and not failing, previously it would fail within a minute.

@cockroach-teamcity
Copy link
Member

This change is Reviewable

Copy link
Collaborator

@stevendanna stevendanna left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This seems reasonable to address the flake.

It does make me wonder how many tests would be subtly wrong if run in multi-node configurations:

> rg '^SET CLUSTER SETTING' | grep "/testdata/" | wc -l
     252
``

@adityamaru
Copy link
Contributor Author

TFTR!

bors r=stevendanna

@craig
Copy link
Contributor

craig bot commented Mar 25, 2022

🕐 Waiting for PR status (Github check) to be set, probably by CI. Bors will automatically try to run when all required PR statuses are set.

@adityamaru
Copy link
Contributor Author

bors r=stevendanna

@craig
Copy link
Contributor

craig bot commented Mar 25, 2022

🕐 Waiting for PR status (Github check) to be set, probably by CI. Bors will automatically try to run when all required PR statuses are set.

@adityamaru
Copy link
Contributor Author

bors r=stevendanna

@craig
Copy link
Contributor

craig bot commented Mar 25, 2022

🕐 Waiting for PR status (Github check) to be set, probably by CI. Bors will automatically try to run when all required PR statuses are set.

@adityamaru
Copy link
Contributor Author

bors r=stevendanna

@craig
Copy link
Contributor

craig bot commented Mar 25, 2022

🕐 Waiting for PR status (Github check) to be set, probably by CI. Bors will automatically try to run when all required PR statuses are set.

@adityamaru
Copy link
Contributor Author

bazel failure is an unrelated flake

bors r=stevendanna

@adityamaru
Copy link
Contributor Author

bors r=stevendanna

craig bot pushed a commit that referenced this pull request Mar 26, 2022
78489: backupccl: set shorter closed_timestamp settings in datadriven tests r=stevendanna a=adityamaru

This change drops the `kv.closed_timestamp.side_transport_interval`
and `kv.closed_timestamp.target_duration` in the servers started as part
of the datadriven framework. This change was motivated by the fact that
cluster settings set in certain multi-node datadriven tests were not being
observed by all nodes in the cluster. Cluster settings are now backed by
rangefeeds instead of gossiping the SystemSpanConfig and so these settings
should speed up when all nodes become aware of changes to their values.

Fixes: #77264

Release note: None

Co-authored-by: Aditya Maru <adityamaru@gmail.com>
@craig
Copy link
Contributor

craig bot commented Mar 26, 2022

Build failed:

@adityamaru
Copy link
Contributor Author

bors r+

@craig
Copy link
Contributor

craig bot commented Mar 26, 2022

Build failed:

This change drops the `kv.closed_timestamp.side_transport_interval`
and `kv.closed_timestamp.target_duration` in the servers started as part
of the datadriven framework. This change was motivated by the fact that
cluster settings set in certain multi-node datadriven tests were not being
observed by all nodes in the cluster. Cluster settings are now backed by
rangefeeds instead of gossiping the SystemSpanConfig and so these settings
should speed up when all nodes become aware of changes to their values.

Fixes: cockroachdb#77264

Release note: None
@adityamaru
Copy link
Contributor Author

Flakes again 😢 Going to see if bors lets me through.

bors r=stevendanna

@craig
Copy link
Contributor

craig bot commented Mar 27, 2022

Build succeeded:

@craig craig bot merged commit 27075f3 into cockroachdb:master Mar 27, 2022
@adityamaru adityamaru deleted the fix-flaky-column-families branch March 28, 2022 07:52
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

ccl/backupccl: TestBackupRestoreDataDriven failed
3 participants