hubcio commented on PR #4454: URL: https://github.com/apache/iggy/pull/4454#issuecomment-6085793321
i don't think we can merge this as is. max-parallel creates the next matrix job only when the previous one finishes, and that job then waits for a runner on its own, so k serial jobs pay the queue k times. this PR's own run (37774161426) took 289 min, not the predicted 28. tests couldn't start before minute 86, then the Other matrix at max-parallel 1 queued 38 + 31 + 30 + 17 + 55 min for five jobs, 172 min of waiting for 30 min of work. peak running concurrency was 9. the same jobs without caps replay to ~91 min. #4479 cut the queue, but this layout multiplies what's left by about 6. on today's runs a 1-3 min queue turns 22-31 min into 32-50, and the 17 min queue at 12:25Z would have turned 41 into 141. maybe we should get under 20 with fewer, fatter jobs instead: one ubuntu-slim job for the 14 sub-minute common checks, one job per sdk, rust tests uncapped and ungated. also the asf text counts running jobs, max-parallel counts queued ones too.. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
