[
https://issues.apache.org/jira/browse/FLINK-20653?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17253970#comment-17253970
]
Yun Tang commented on FLINK-20653:
----------------------------------
Though the code of checkpoint store changed much recently, I think the logic is
still storing metadata first and then clean up the unreferenced incremental
state:
Code below is pasted from
[DefaultCompletedCheckpointStore.java#addCheckpoint.|https://github.com/apache/flink/blob/release-1.12/flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/DefaultCompletedCheckpointStore.java#L201-L226]
And as you can see, Flink stores the metadata in
{{checkpointStateHandleStore.addAndLock(path, checkpoint)}} and then discard
previous unreferenced state in {{tryRemoveCompletedCheckpoint}} if no exception
meet.
{code:java}
public void addCheckpoint(
final CompletedCheckpoint checkpoint,
CheckpointsCleaner checkpointsCleaner,
Runnable postCleanup) throws Exception {
checkNotNull(checkpoint, "Checkpoint");
final String path =
completedCheckpointStoreUtil.checkpointIDToName(checkpoint.getCheckpointID());
// Now add the new one. If it fails, we don't want to loose
existing data.
checkpointStateHandleStore.addAndLock(path, checkpoint);
completedCheckpoints.addLast(checkpoint);
// Everything worked, let's remove a previous checkpoint if
necessary.
while (completedCheckpoints.size() >
maxNumberOfCheckpointsToRetain) {
final CompletedCheckpoint completedCheckpoint =
completedCheckpoints.removeFirst();
tryRemoveCompletedCheckpoint(
completedCheckpoint,
completedCheckpoint.shouldBeDiscardedOnSubsume(),
checkpointsCleaner,
postCleanup);
}
LOG.debug("Added {} to {}.", checkpoint, path);
}
{code}
I have to say this part of code is refactored in release-1.12, and could you
please share your exception logs with DFS audit log in detail to see whether
there existed something unexpected leading data lost?
> SharedState should be registered after PendingCheckpoint storeCheckpoint
> ------------------------------------------------------------------------
>
> Key: FLINK-20653
> URL: https://issues.apache.org/jira/browse/FLINK-20653
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Checkpointing
> Affects Versions: 1.12.0
> Reporter: weiyunqing
> Priority: Major
> Attachments: 123.png, image-2020-12-17-20-20-21-246.png
>
>
> SharedState should be registered after PendingCheckpoint storeCheckpoint，If
> the metadata storage fails, the deleted incremental state cannot be recovered。
> !image-2020-12-17-20-20-21-246.png!
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
