ci: replace single gating check with wait-for-status-check [no-ci]#1008
ci: replace single gating check with wait-for-status-check [no-ci]#1008leofang merged 1 commit intoNVIDIA:mainfrom
wait-for-status-check [no-ci]#1008Conversation
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
b455cfe to
7a583d3
Compare
7a583d3 to
6b6844a
Compare
|
/ok to test 6563550 |
This comment has been minimized.
This comment has been minimized.
6563550 to
d1ec467
Compare
|
/ok to test |
|
Auto-sync is disabled for ready for review pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
|
/ok to test |
|
Check job still completes instead of spinning: https://github.com/NVIDIA/cuda-python/actions/runs/18165463144/job/51706593048?pr=1008 |
|
@leofang Can you paste the names of the checks here from the UI? I can't see them because I don't have access. |
|
Once we are sure the check job completes the last, we can either adjust the repo setting and then merge, or admin-merge then adjust the repo setting. But we are not there yet. The repo setting is irrelevant of the failure of waiting for dependent jobs here. It is just a workflow bug that we should fix first. |
|
/ok to test |
1 similar comment
|
/ok to test |
|
I'm going to try one more alternative to avoid the action entirely. |
|
I really don't like having to add an arbitrary delay to get this to work. |
|
Yeah adding a random delay is flakey. It could proceed to complete when build jobs finish but the test jobs are still being spawned. Not a fan either. |
|
/ok to test |
3 similar comments
|
/ok to test |
|
/ok to test |
|
/ok to test |
wait-for-status-checkwait-for-status-check [skip ci]
|
/ok to test |
I applied it. You can see it now becomes "required". I haven't removed the old check yet. But I am confused: If this is required, we still need to execute |
|
The new status check job should be considered passing, because it's skipped, according to that documentation. I'm try to taking advantage of the fact that skip == success. |
wait-for-status-check [skip ci]wait-for-status-check [no-ci]
|
/ok to test |
8203214 to
1685f32
Compare
|
/ok to test |
|
@leofang It looks like the |
|
I am checking. This is odd because as discussed earlier once a CI job is run, it should appear as a registered job in the repo setting's dropdown menu, but for some reason I had to type the full name explicitly without anything popping out for me to choose. |
|
@cpcloud I think I fixed it. It's just "Check job status" (without "CI"), see the last row: |
Yep, that has to be removed before merge is allowed. |
|
Let me check quickly. btw we'll need to backport this PR to 12.9.x. Keith applied the same ruleset to all branches. |
|
It works now! To avoid race condition I am merging immediately. |
|
Thanks for the review, much appreciated! |
|
Thanks for working this out Phillip! |
|


This PR changes our usage of
status-check.ymlto allow skipping CI when the head commit of PR contains the text[skip ci].Our current usage of the
wait-for-checkaction doesn't sense to me.We have as inputs all the checks that we care about
build,test*,docs,and we only invoke that action when those succeed.
That seems to be no different than running
exit 0againstubuntu-latestandthus defeats the purpose of using
wait-for-checkswhich is to allowfiner-grained control over what is actually a required check.
In our case, there are cases such as #999 where it doesn't make sense to do
a full build test docs run, but developers are left waiting on an admin to
merge their PR.
Proposal for merging this PR:
I don't have permissions to do that, so someone other than me needs to do
that.