Build: cancel a build when a new one for the same branch/PR is triggered #8961
Comments
When the exception is an exception we know it does not require communicating the VCS service about its state, we avoid sending the build status to the VCS service. This is currently used to avoid sending a "failure" (red cross) notification when the build is marked as duplicated. In the future, it could also be used when the build is auto-cancelled because a new commit on the same branch was done (see #8961) Closes #9040
When the exception is an exception we know it does not require communicating the VCS service about its state, we avoid sending the build status to the VCS service. This is currently used to avoid sending a "failure" (red cross) notification when the build is marked as duplicated. In the future, it could also be used when the build is auto-cancelled because a new commit on the same branch was done (see #8961) Closes #9040
|
that would be a really nice feature :) |
|
This would help us over at https://github.com/xgcm/xgcm quite a bit. +1 👍 |
|
I think the SunPy and Astropy folks would love having this too :) @nabobalis @pllim |
I cannot speak for all of Astropy but personally I would love auto-cancel. Otherwise, sometimes when we get commit spam, RTD gives us a long queue. Thanks for the ping, @astrojuanlu ! |
|
Yeah from the SunPy side, we waste so much of your CI time (with how we commit) and implementing this would help. I try to cancel builds manually but I don't keep on top of this. |
|
Also chiming in that this would be a welcome addition for our PRs at https://github.com/Ouranosinc/xclim |
This commit implements a simple logic to cancel old running builds when a new build for the same project/version arrives: 1. look for running builds for the same project/version 2. if there are any, it cancels them all one by one via Celery's revoke method 3. trigger a new build for the current commit received Note that this new feature is behind a feature flag (`CANCEL_OLD_BUILDS`) for now so we can start testing it on some projects that have shown their interest on this feature. The current behavior for `DEDUPLICATE_BUILDS` will be replaced by this new logic in the future. However, it was not removed in this commit since it's still useful for projects that won't be using the new feature flag yet. Closes #8961
|
I went ahead and opened #9549 as a starting point. Let's see if we can have it reviewed and merged soon to start testing this new feature. |
|
I'm targeting this feature to go out tomorrow. Once deployed, I will enable this feature on the following projects:
Let me know if there are other people subscribed to this issue that want to try this feature at this beta phase and I will enable it for their projects as well. Hopefully, after some feedback from you 🙏🏼 , we can enable it by default 👍🏼 . Thank you all! |
|
This is currently deployed and all the projects listed in my previous comment have the feature enabled already. I quickly QAed this on production and it did work fine. Please, open an issue if you find something that's not working as you expected or contact support if you want to provide feedback to us! 💪🏼 |
|
It seems to be doing the job:
Thanks again for this! |
|
Hello. I think it works but I keep forgetting to check. Can you please enable it for this also? This one has less contributors to keep track of and I think we push subsequent commits pretty frequently there, so it is easier for me to confirm. Thanks! 🙏 |
|
@pllim Thanks for your feedback. I enabled it on https://jdaviz.readthedocs.io/en/latest/ which I think is the Read the Docs project linked to that GH repository. My main concern here is to not cancel valid builds. I mean, if we are not always cancelling a build that should be cancelled it's not a big problem. However, if we are cancelling a build that shouldn't be cancelled that's a problem. In theory, this has been working fine since we enabled it. We haven't received any complain yet and the tests I've personally done did work. However, I suppose that projects doing many small pushes with one commit and multiple collaborators may have better feedback than myself. Let me know if you think it works as you expected. Thanks again for your feedback. BTW, we plan to enable it for all the projects in the following weeks if we don't receive any complain 😄 |
|
Haven't noticed any problems. Thanks again! 🙇♀️ |
Now that we implemented a way to revoke Celery tasks we are able to cancel a running build. This is useful for people to manually cancel a build they are not interested in finishing for different reasons and free up the concurrency limit they have.
There are some scenarios where we can handle this automatically because we already know that the running build is useless and can be canceled:
Related #8850
The text was updated successfully, but these errors were encountered: