Github user jinxing64 commented on the issue: https://github.com/apache/spark/pull/18388 Analyzing the heap dump, there are 200K+ connections and 3.5M blocks(`FileSegmentManagedBuffer`) being fetched. Yes, flow control is a good idea. But I still think it make much sense to control the concurrency. Reject some "open blocks" requests, thus we can have sufficient bandwidth for the existing connections and we can finish the reduce task as soon as possible. Simple flow control(slow down connections when pressure) can help avoid OOM, but it seems more reduce tasks will run longer.
--- 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