rmetzger commented on pull request #15071: URL: https://github.com/apache/flink/pull/15071#issuecomment-790590093
Thanks a lot for your review! You had very helpful comments (because I clearly had some mistakes in there). I'm looking forward to further feedback. - I've put the parallelism change logic into the JobMaster, because there I have access to the JobGraph and the scheduler factory (which can tell me which scheduler we are using). - I couldn't think of any checks for programmatically requiring in the AdaptiveScheduler that the JobGraph is already properly configured. - I'm undecided if my approach in the ITCase with the CountDownLatch is intuitive. I didn't want to spend more time thinking about alternatives (guarded integer, lock object). I would appreciate a quick comment about it (yay or nay). I would also like to mention that the ITCase currently executes quite slowly, because scale ups only happen after 10 seconds. This will be fixed as part of FLINK-21136. Also, the ITCase configuration currently requires setting adaptive scheduler and reactive scheduler mode. Once adaptive scheduler is automatically set on reactive mode, I'll simplify the test configuration. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
