[ https://issues.apache.org/jira/browse/FLINK-5940?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15890631#comment-15890631 ]
ASF GitHub Bot commented on FLINK-5940: --------------------------------------- GitHub user tillrohrmann opened a pull request: https://github.com/apache/flink/pull/3451 [backport-1.1] [FLINK-5940] [checkpoint] Harden ZooKeeperCompletedCheckpointStore.recover method Backpor of #3446 onto `release-1.1` branch. The ZooKeeperCompletedCheckpointStore only tries to recover the latest completed checkpoint even though it might have read older checkpoint state handles from ZooKeeper. In order to deal with corrupted state handles, this commit changes the behaviour such that the completed checkpoint store adds all read retrievable state handles from ZooKeeper and upon request of the latest checkpoint it will return the latest completed checkpoint which could be retrieved from the state handles. Broken state handles are removed from the completed checkpoint store and ZooKeeper. You can merge this pull request into a Git repository by running: $ git pull https://github.com/tillrohrmann/flink fixCheckpointRecoveryBp1.1 Alternatively you can review and apply these changes as the patch at: https://github.com/apache/flink/pull/3451.patch To close this pull request, make a commit to your master/trunk branch with (at least) the following in the commit message: This closes #3451 ---- ---- > ZooKeeperCompletedCheckpointStore cannot handle broken state handles > -------------------------------------------------------------------- > > Key: FLINK-5940 > URL: https://issues.apache.org/jira/browse/FLINK-5940 > Project: Flink > Issue Type: Bug > Components: State Backends, Checkpointing > Affects Versions: 1.2.0, 1.1.4, 1.3.0 > Reporter: Till Rohrmann > Assignee: Till Rohrmann > > The {{ZooKeeperCompletedCheckpointStore}} reads a set of > {{RetrievableStateHandles}} from ZooKeeper upon recovery. It then tries to > retrieve the {{CompletedCheckpoint}} from the latest state handle. If the > retrieve operation fails, then the whole recovery of completed checkpoints > fails even though the store might have read older state handles from > ZooKeeper. > I propose to harden the behaviour by removing broken state handles and > returning the first successfully retrieved {{CompletedCheckpoint}}. -- This message was sent by Atlassian JIRA (v6.3.15#6346)