[jira] [Commented] (BEAM-12672) Update all Precommits to rerun failed tests K times.

["Valentyn Tymofieiev (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/BEAM-12672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17418893#comment-17418893 ] 

Valentyn Tymofieiev commented on BEAM-12672:
--------------------------------------------

> Note: This does not mean that we should ignore flakey tests, which are sometimes indicative of bugs. We should continue to investigate those with a continuous test suite.

There is a high likelihood that follow up investigation won't be happening if automatic retries are enabled, so i'd be cautious with this until we have the system to investigate and eliminate flakes that works. 

Prior related discussions:

https://lists.apache.org/thread.html/r906708da1dda53fca20498a95dda4913d80ce453a298359dd38e9b31%40%3Cdev.beam.apache.org%3E
https://lists.apache.org/thread.html/16060fd7f4d408857a5e4a2598cc96ebac0f744b65bf4699001350af%40%3Cdev.beam.apache.org%3E

> Update all Precommits to rerun failed tests K times.
> ----------------------------------------------------
>
>                 Key: BEAM-12672
>                 URL: https://issues.apache.org/jira/browse/BEAM-12672
>             Project: Beam
>          Issue Type: Test
>          Components: testing
>            Reporter: Alex Amato
>            Priority: P2
>
> We should configure the precommit test suites to re-run failed tests a few (K) times.
> (A bonus feature would be to tag certain tests as known flakey, and run multiple attempts in parallel when the test is first ran)
> The benefits of this are to:
>  * Not have flakey tests slow down PR reviews and merges.
>  * And avoiding the need to repeatedly manually rerun the precommit test with a PR comment.
>  * Getting a clear signal that failed tests are due to the code in the PR
>  * Save computing resources (Rather than needing to run the entire test suite every time it fails, only the failed tests will rerun).
> (Its not realistic to have 0 flakeyness, so this is desirable to improve productivity, as it can take several attempts to have all the tests pass in a single run).
> Note: This does not mean that we should ignore flakey tests, which are sometimes indicative of bugs. We should continue to investigate those with a continuous test suite.
>  



--
This message was sent by Atlassian Jira
(v8.3.4#803005)