-
Notifications
You must be signed in to change notification settings - Fork 17
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Reduce CI runs #184
Reduce CI runs #184
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Will also apply fail-fast: true to the test jobs once we've finished discussions on the matrix.
We discussed offline and the new matrix reflects the consensus @ajschmidt8 @bdice and I came to. |
The latest run on the cudf test PR verifies that this workflow is valid: https://github.com/rapidsai/cudf/actions/runs/8116036240?pr=15201 |
New run based on setting v100/a100 appropriately: https://github.com/rapidsai/cudf/actions/runs/8116093123?pr=15201 |
Net changes to PRs: amd64 jobs go from 8 (3 conda C++, 3 conda Python, 2 wheel) to 5 (2 conda C++, 2 conda Python, 1 wheel) Of course, we also have to account for custom jobs like docs and notebooks that also consume amd64 runners. That usually adds about 2 jobs to the amd64 runners, depending on the repository. For this to improve CI times, we have to be waiting on amd64 (not arm64). That's currently true, from my observations, and I expect that reducing amd64 usage will be sufficient to improve CI queue times. If you also consider that one of those amd64 jobs uses the "earliest" driver and thus comes from a separate pool of runners, this is an even more significant improvement. 👍 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
found two invalid combinations
#184 (comment) Confirming the earlier hypothesis about imbalanced queues for latest/earliest drivers -- for this job's C++ tests, the amd64 "latest" job is still waiting on a runner after 45 minutes, but the amd64 "earliest" job started immediately. |
Aside from some unrelated failures, the test cudf job looks good now. @ajschmidt8 I think this is good to go when you're happy with it. |
This PR cuts down on the the number testing jobs in PRs while attempting to retain similar coverage. It also enables
fail-fast
so that matrix jobs will not proceed to completion if any one of the tests fail.Contributes to rapidsai/build-planning#5