Github user tgravescs commented on the issue: https://github.com/apache/spark/pull/17113 so unfortunately I haven't actually been seeing this. You can see with external shuffle is something happens to the NM and it does cause job failure. NM crashes for OOM, something else kills it accidentally, etc. In those cases the executors are still running fine and map tasks won't fail if its already registered, then you get fetch failure again and iterate until job failure. An alternative to this might be to have the executor re-ping the external shuffle service before starting a new task and have it fail at that point. We could also add in logic to only blacklist affect certain number of failures, but I don't see this as much different as SPARK-19753 where its going to invalidate immediately.
--- 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