potiuk commented on pull request #19499: URL: https://github.com/apache/airflow/pull/19499#issuecomment-965146267
Yeah. It's double-edged sword. Often self-hosted are usually faster than the public runners. However @ashb preparing to the buidl meeting tomorrow, I was looking at the latest charts which I added here: https://cwiki.apache.org/confluence/display/INFRA/Builds+Agenda+2021-11-11 it looks like GitHub Actions implemented some more "fair" algorithm of distributing public runners to the projects in the meantime. Seems that there are no mores peaks of ~500 queued jobs from one project (which was the main reason our 150 job queue was stuck for literally hours. And I have not seen many CI builds/PRs from users waiting for a looong time. What I saw is what @khalidmammadov observed - that because we have limited queue of self-hosted, the regular PRs timed out because their `build-image` workflow did not have a chance to complete due to many committers pushing their PRs. I think maybe it actually makes sense to bring the build-image jobs from regular PRs to use public runners ? At lest to try. In VAST majority of cases those are 1 - 2 minute jobs and all of them are non-parallel (docker builds do not parallelise well) so they won't be much slower. WDYT @ashb ? -- 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]
