You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by uce <gi...@git.apache.org> on 2017/10/20 10:11:22 UTC
[GitHub] flink issue #4254: [FLINK-7067] [jobmanager] Fix side effects after failed c...
Github user uce commented on the issue:
https://github.com/apache/flink/pull/4254
@tillrohrmann Thanks for looking over this. The `TestingCluster` is definitely preferable. I don't recall how I ended up with the custom setup instead of the `TestingCluster`.
I changed the test to wait for another checkpoint after the failed savepoint. I also considered this for the initial PR, but went with mocking in order to test the case that periodic checkpoints were not activated before the cancellation [1]. I think the current variant is a good compromise between completeness and simplicity though.
[1] As seen in the diff of `JobManager.scala`, we only activate the periodic scheduler after a failed cancellation iff it was activated before cancellation. This case can't be tested robustly with the current approach. We could wait for some time and if no checkpoint arrives in that time consider checkpoints as not accidentally activated, but that's not robust. I would therefore ignore this case if you don't have another idea.
---