[FLINK-9815][yarn][tests] Harden tests against slow job shutdowns #6352
Conversation
zentol
changed the title
|
the test failures may highlight tests that weren't shutting down the last application properly; previously this would've succeeded since the check was done in |
|
will look into them tomorrow |
There was a problem hiding this comment.
Thanks for improving the YarnTestBase @zentol. The changes look good to me. Before merging, it would be good to fix the failing test cases.
| if (runningApp.isPresent()) { | ||
| final ApplicationReport app = runningApp.get(); | ||
| Assert.fail("There is at least one application on the cluster that is not finished." + | ||
| "App " + app.getApplicationId() + " is in state " + app.getYarnApplicationState()); |
There was a problem hiding this comment.
Could we log all running applications instead of any?
|
The failing test case seems to be caused by |
|
The |
|
Interesting. I thought I would have executed |
|
yay travis is green. |
What is the purpose of the change
This PR hardens the
YarnTestBaseagainst jobs that just don't want to shut down that quickly (i.e. within 500ms).The maximum waiting time has been increased to 10 seconds, during which we periodically check the state of all applications.
Additionally, the failure condition from
@Beforewas moved to the@Aftermethod.This change will allow us to better differentiate between simple timing issues and unsuccessful job shutdowns.