Builds: cancelled build left in triggered state

[agjohnson]

Not sure of the cause here, but this is one of the invalid states that builds can end up in: cancelled but still in a triggered state.

Here is what logging shows:

• Build has healthcheck=None, it was never started by a worker
• Build was stuck retrying
• User cancelled build
• Build task is still retried after cancellation

https://onenr.io/0Bj3J0JydRX

It looks like task revocation does nothing in this case.

[humitos]

Do you know how to reproduce this?

I tried triggering a build for the same version twice and the first build was cancelled automatically successfully. I also tried cancelling the build manually and it was cancelled successfully too.

• https://app.readthedocs.org/projects/test-builds/builds/29769057/
• https://app.readthedocs.org/projects/test-builds/builds/29769062/

I also hit concurrency build, the build was scheduled to be retried in 5 minutes and I cancelled it immediately. Worked:

• https://app.readthedocs.org/projects/test-builds/builds/29769112/

[agjohnson]

I don't, you'll have to look at the logs for this one. That project was creating builds in this state fairly commonly. Some of the recent changes to the build task may have helped avoid this scenario too though.

[humitos]

I'm transferring this issue to readthedocs-builder. However, we shouldn't be able to reproduce it there because how the new logic works. I want to double check this once we are migrated, tho.