[8.x] Improve task manager functional tests in preperation for mget task claimer being the default (#196399) #197062
Conversation
…aimer being the default (elastic#196399) Resolves elastic#184942 Resolves elastic#192023 Resolves elastic#195573 In this PR, I'm improving the flakiness found in our functional tests in preperation for mget being the default task claimer that all these tests run with (elastic#194625). Because the mget task claimer works differently and also polls more frequently, we end-up in situations where tasks run faster than they were with update_by_query, creating more race conditions that are now fixed in this PR. Issues were surfaced via elastic#190148 where I set `mget` as the default task claiming strategy. Flaky test runs (some of these failed on other tests that are flaky): - https://buildkite.com/elastic/kibana-flaky-test-suite-runner/builds/7151 - https://buildkite.com/elastic/kibana-flaky-test-suite-runner/builds/7169 - https://buildkite.com/elastic/kibana-flaky-test-suite-runner/builds/7172 - https://buildkite.com/elastic/kibana-flaky-test-suite-runner/builds/7175 - https://buildkite.com/elastic/kibana-flaky-test-suite-runner/builds/7176 - https://buildkite.com/elastic/kibana-flaky-test-suite-runner/builds/7185 (for elastic@0fcf1ae) (cherry picked from commit 3b8cf12) # Conflicts: # x-pack/test/alerting_api_integration/spaces_only/tests/alerting/group4/alerts_as_data/alerts_as_data_flapping.ts
💚 Build Succeeded
Metrics [docs]
|
| OneOfTaskTypes('task.taskType', claimPartitions.unlimitedTypes), | ||
| OneOfTaskTypes( | ||
| 'task.taskType', | ||
| claimPartitions.unlimitedTypes.concat(Array.from(removedTypes)) |
There was a problem hiding this comment.
Is this going to cause a starvation problem? I don't think so, as it looks like we are removing them (marking them to unrecognized) at the ed of the claim, but not clear to me why this wasn't in here before. Perhaps there was some confusion since we also have an issue to handle these via a separate stand-alone task?
This code does seem right, so I'm guessing some test was expecting these to be marked as unrecognized, and perhaps that wasn't happening before this PR ...
There was a problem hiding this comment.
I'm guessing some test was expecting these to be marked as unrecognized, and perhaps that wasn't happening before this PR ...
Yeah there was a test in the update_by_query suite that ensured removed task types get marked as unrecognized:
I was able to make the test pass by adding this bit to the mget filter but I agree we should move this to its own task next via #192686
Backport
This will backport the following commits from
mainto8.x:Questions ?
Please refer to the Backport tool documentation