Github user uce commented on a diff in the pull request:
https://github.com/apache/flink/pull/1633#discussion_r52986635
--- Diff:
flink-runtime/src/main/scala/org/apache/flink/runtime/jobmanager/JobManager.scala
---
@@ -1073,57 +1073,73 @@ class JobManager(
// execute the recovery/writing the jobGraph into the
SubmittedJobGraphStore asynchronously
// because it is a blocking operation
future {
- try {
- if (isRecovery) {
- executionGraph.restoreLatestCheckpointedState()
- }
- else {
- val snapshotSettings = jobGraph.getSnapshotSettings
- if (snapshotSettings != null) {
- val savepointPath = snapshotSettings.getSavepointPath()
+ val restoreStateSuccess =
+ try {
+ if (isRecovery) {
+ executionGraph.restoreLatestCheckpointedState()
--- End diff --
Yes, it had the same issue as the savepoint restore. If restoring the
checkpoint failed, job submission was never ACK'd, but execution was restarted
etc. I'm not sure that this is the behaviour we want in the long term, but I
think failing the submission is clearer behaviour right now.
Another issue I've noticed is the following: when HA checkpoint recovery
fails and the job cannot be scheduled for execution, the job will eventually be
removed from ZooKeeper and hence never be recovered again. Let me open an issue
for this...
---
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 [email protected] or file a JIRA ticket
with INFRA.
---
