Github user squito commented on the issue: https://github.com/apache/spark/pull/18874 @srowen you have a good point about a case that becomes worse after this change. Still I think this change is better on balance. btw, there are more even more odd cases with dynamic allocation right now -- the one that I've seen most often is if you have a run really short tasks, but all in sequence, you probably won't release any executors. Say you first run some really large job with 10k tasks, and so you request a bunch of executors. After that, you only ever run 100 tasks at a time, so you could release a bunch of resources. But in 60 seconds, its enough time for a bunch of short stages to execute. Each stage chooses a random set of executors to run on. So no executor is ever idle for 60 seconds. Perhaps we could also fix that in some larger change which more tightly integrated dynamic allocation into the scheduler that Tom was alluding to.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. --- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org