Open weikanglim opened 1 month ago
In this particular case the assumption was made that the GitHub token would not be rate limited. A recent test of some automation broke this assumption but that should not be the case going forward.
A few ways to solve this would be to:
publish-cli.yml
in some cases because it's not strictly necessary (we shouldn't do this) I'm closing this since it was caused by a one-off event but if it happens again we'd probably start with ensuring that automation tests use a different token.
@danieljurek Is there a property for re-uploading in-place? I feel like this would be acceptable for a non-production artifact.
Btw: I hit this twice again today. Here's another instance: https://dev.azure.com/azure-sdk/internal/_build/results?buildId=4172035&view=results
Another tactical thing: can we shift the PR add comment up? (haven't looked at the implementation here)
Reopen! This is still a problem let's do something about it.
Is there a general strategy we can make artifact publishing idempotent to job re-runs?
Example:
In publish-cli.yml (and anywhere else), when we have a pipeline artifact being published:
And in a subsequent step such as create-pull-request, fails due to a flaky issue. On a rerun, we will get an error:
Here's a recent build run demonstrating this issue.