Github user colorant commented on the pull request: https://github.com/apache/spark/pull/1328#issuecomment-48688983 I think there might be several issues need to be addressed here to make this more robust and sound solution: 1. A solution to avoid pre-start stage occupy too many CPU resource which starve the parent stage and prevent it from finish in time. will need to control how many pre-start task can launch and adjust according to parent stage's status. And need some config to allow user to fine tune the behavior, after all, what's the best time to pre-start might depends on actual case. 2. might need a more common Interface, seems a lot of specific class implementation is assumed. 3. Might need to take more care of the memory management issue, say how to prevent pre-start stage occupy too many memory (say by cache etc) which complicate the overall cache eviction and GC problem etc. Just my suggestion.
--- 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. ---