Github user carsonwang commented on the pull request: https://github.com/apache/spark/pull/8090#issuecomment-129724889 @squito , here are some further information about the issue. I was running an iterative graph-parallel algorithm which uses RDD cache for iterative computations. In job 3, if everything works correctly, only stage 16 and stage 17 will run, other stages 12-15 are skipped stages. In the failure case, stage 17 failed due to a fetch failure from stage 16. So the scheduler resubmitted stage 16 which in turn submitted the missing stage 12. Stage 12 reads data from HDFS but it only computes part of the partitions instead of all the partitions. In this case, the internal accumulators are not created and it causes an exception when the accumulators are accessed later.
--- 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