rdblue commented on issue #28129: [SPARK-31346][SQL]Add new configuration to make sure temporary directory cleaned URL: https://github.com/apache/spark/pull/28129#issuecomment-609925222 It seems like this is trying to avoid state building up, but the thread that's doing the delete may get killed due to speculation or other causes. Rather than choosing between cleaning up those accumulating paths (the current behavior) and always cleaning on exit (the new behavior), it would be nice to have a third option. What I'm thinking of is an option where the path is put into a separate queue to be cleaned up. That way, killing the task won't stop the cleanup but something is still responsible for the delete and removing from the `deleteOnExit` state. That said, I think this is a reasonable change for some environments. We run in YARN with dynamic allocation, so our executors are short-lived enough that I would just turn this on by default.
---------------------------------------------------------------- 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] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
