So the async poll operations should use different tmpdirs from the command that launched the job so this shouldn’t happen, at least in theory. If the same tempdir is in fact used for subsequent polling calls we’d want to fix that – but I don’t think that’s the case (haven’t seen that happen anyway). Delete also happens toward the end. It may be a corner case where yum has some startup lag and it’s polling before the results file is written yet – and as a result what async_status is looking for isn’t there yet – but that saying “Arguments file” would lead me to think otherwise.
You may be able to add some extra debug to the async_status module to find out what is going on.
OT – there should be no need to make those two operations async. Running a command to “yum update all”, sure, but IMHO the yum module should always be used non-async and it would be more efficient to do it that way. We should still find out how to fix this, of course.