Closed hendrikmakait closed 4 weeks ago
_See test report for an extended history of previous test failures. This is useful for diagnosing flaky tests._
28 files - 1 28 suites - 1 10h 51m 38s :stopwatch: + 52m 45s 4 053 tests - 4 3 950 :white_check_mark: + 10 97 :zzz: - 9 6 :x: - 1 54 399 runs +6 165 52 292 :white_check_mark: +5 997 2 099 :zzz: +185 8 :x: ±0
For more details on these failures, see this check.
Results for commit 2c26a72d. ± Comparison against base commit b3f8fcd0.
:recycle: This comment has been updated with latest results.
@hendrikmakait who around do you think would be good to review more deeply here? Do you want me to take a longer look? Maybe Alex has knowledge here?
Alex might have some knowledge here, but he's out today. We have reasonable test coverage for rechunking, so a careful sanity check should suffice here. (In fact, it caught an issue I had introduced early in creating this fix.) @wence- would probably be a good fit, but I'm not sure whether he has any capacity.
I'm currently running an A/B test as well to compare performance.
Maybe I'm saying "let me know if I my getting engaged would meaningfully accelerate things"
Maybe I'm saying "let me know if I my getting engaged would meaningfully accelerate things"
If you have the time to review this today and approve it, that's the quickest way of getting this PR merged. From previous interactions, it sounded like Alex hasn't dug deeply into P2P before, so he probably doesn't know more than you here.
I'm still waiting for A/B test results from this run: https://github.com/coiled/benchmarks/actions/runs/9257990760/job/25467247783
Thanks for the review, @wence-!
Closes #8656
Previously, independent partials of a P2P rechunk would create their own independent task groups. This could lead to an explosion of the task group count and slow down the scheduler. This PR fixes that.
pre-commit run --all-files