-1 for reason @pingou gave on list: build and tag are separate operations. We can fix the tagging on a build that succeeded but where the automatic tag failed for some reason. There are probably other ways we can improve this situation (like making it easier for the packager to trigger an attempt to fix the tagging), but treating the build as failed seems wrong to me.

Maybe the best of two worlds was koji mark build as
failed but let resubmit the failed task and completed the build
successfully .

koji builds marked as completed but with fails tasks should be avoided,
in my opinion.

This particular error is likely the result of a retried hub call or a restarted build task. The COMPLETE state for the build indicates that a build process was able to call host.completeBuild. So the build entry was done. Something went wrong in the task handler after that point.

Unfortunately something wonky happened with the task. I suspect that the task was restarted at just the wrong time (in the very brief span between completeBuild and the end of the task). The following pass happily noted the already finished subtasks, but failed on the duplicate completeBuild call.

Failed tasks for successful builds are rare, but they do happen. They are almost always interesting cases worth investigation. However, we should not retroactively fail a build once it is marked completed.

JFYI , this happened in a mass rebuild when automatic update failed on Bodhi [1] and also happened to me when I sent one koji cancel , after already built the package, but before tag it [2].

[1]
Result: GenericError: cannot update build 1424799, state: COMPLETE

[2]
koji cancel 62407047

This issue has been migrated to Fedora Forge:
https://forge.fedoraproject.org/koji/koji/issues/1895

Please continue any further discussion there.