kubeflow / tf-operator / 870

71%

push

travis-ci

Fix bug with jobs not working if you recreate a job with same name as previous job (#399)

This appears to be a bug with the informer forgetting about active jobs.

I think the root cause is that delete events are processed directly
as opposed to using the work queue. This causes race conditions because
the work queue is what ensures only on thread is processing a given job
at a time.

Since we no longer explicitly handle delete events we rely on garbage
collection and owner references to delete the pods.

We add testing to our test runner to verify pods are garbage collected.

We update test_runner.py to try recreating jobs with the same name.
In general this doesn't reproduce the original bug.

Related to #322 (to need to push a new image to fix).
Fix #310

Related to #373

3 of 3 new or added lines in 1 file covered. (100.0%)

580 of 1820 relevant lines covered (31.87%)

2.65 hits per line