Github user rdblue commented on the issue: https://github.com/apache/spark/pull/13482 @tgravescs, removing `notifyAll` doesn't solve the problem entirely, it just removes one path that's causing the `allocate` call to be run too many times. (Also, I haven't tested delaying loss reasons in our Spark jobs at scale, other than for the 200ms introduced here.) Ensuring that `allocate` is not called too often addresses the problem no matter what the immediate cause is. That's why I think it's a good idea to fix the two separately: first ensure that `allocate` will not run too often and starve other operations on the `YarnAllocator` and, second, track down the cases that cause this. Even if we were to fix the `YarnAllocator` so we don't have resource contention, ensuring a min interval between calls to `allocate` is a good idea so that Spark doesn't make too many useless calls to the resource manager. And, I don't want to track down this same bug in 3 months because of a different path trigger it.
--- 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