Github user kanzhang commented on the pull request: https://github.com/apache/spark/pull/1338#issuecomment-50210598 > Even if we cached the results of the read, I think we'd still need to keep a copy of the configuration in case any of those cached partitions are lost and we need to recompute from scratch. This is where I have a gap. I imagined the error recovery code would have kept a copy of the conf on driver and re-broadcast it when re-computation happens, whereas those broadcasted copies on executors could have been cleaned after job is successfully run the first time.
--- 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. ---