Closed dalcinl closed 11 months ago
@conda-forge-admin, please rerender
@isuruf @ocenpaf @minrk Guys, I need help here. After merging this PR, all the built packages failed to upload.
re-running in case the failed upload was an intermittent error
Please stop randomly smashing buttons. You've managed to restart the exact same job ~20 times which leads to complete resource starvation in conda-forge:
I hit literally one button to restart the job. You can blame Azure's horrible GitHub integration if it did a bunch of things instead of what was requested.
@h-vetinari, let's give the maintainers some benefit of the doubt and not come into conclusions. It's a known Azure bug.
Guys, would it make sense to make a new PR with let say an empty commit, or a bump of the build number?
@dalcinl, the packages should be uploaded now. We cancelled all of the duplicated jobs. (Azure sometimes restarts n builds with n jobs each instead of just 1 build with n
jobs.)
Thanks! I'll proceed to get all the other PETSc-related packages in.
We can do something like that that next time. This time worked, but it wasted a bunch of resources by starting several duplicate jobs on Azure from a single restart request.
It's a known Azure bug.
I restart bots a lot, and with the following guidelines, it has never happened to me:
Having >20 restarts just seemed like an absurdly high number, hence my reaction. I apologise for jumping to conclusions. Still, please try to follow the above guidelines, and don't click more than once.
FWIW, I'm 100% in favour of giving feedstock maintainers more fine-grained control over how azure jobs are restarted/cancelled, but that's a kind of built-in limitation.
No worries, sorry for wasting resources!
let the job run fully to completion before clicking "re-run", even if that means waiting for the 6h timeout
The last run completed 3 days ago
please try to follow the above guidelines, and don't click more than once.
I will continue to do that, like I did today.
Having >20 restarts just seemed like an absurdly high number,
I believe it creates one job per matrix entry when this happens, and petsc has 20 variants.
FWIW, I'm 100% in favour of giving feedstock maintainers more fine-grained control over how azure jobs are restarted/cancelled, but that's a kind of built-in limitation.
Yeah, I really wish I could use the Azure UI directly instead of the GitHub Checks UI which is such a mess. It's weird that the CI service with by far the worst GitHub integration is the one owned by the same company. I don't think I have permission to cancel any jobs once they are launched.
I don't know if azure has github-actions-style concurrency/cancellation to automatically cancel superseded jobs?
I don't know if azure has github-actions-style concurrency/cancellation to automatically cancel superseded jobs?
Generally this works for pushes to the same PR, but not in cases like this one where a wild array of runs gets triggered.
It's weird that the CI service with by far the worst GitHub integration is the one owned by the same company.
💯
Checklist
0
(if the version changed)conda-smithy
(Use the phrase code>@<space/conda-forge-admin, please rerender in a comment in this PR for automated rerendering)