Github user tillrohrmann commented on the pull request: https://github.com/apache/flink/pull/640#issuecomment-97818576 The backtracking looks good @mxm. I have some remarks concerning the way the locking of the partition on the TMs works. At the moment this happens sequentially, meaning that for each `IntermediateResultPartition` a message is sent to the TM and then the response is awaited. Only after receiving the response of this TM, the next `IntermediateResultPartition` is processed. This can considerably slow down the scheduling if the degree of parallelism is high. I think we should make use of Akka's future composition to do that concurrently. Furthermore, we could think about doing the backtracking in parallel. This could also speed up the scheduling process.
--- 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. ---