[
https://issues.apache.org/jira/browse/FLINK-5667?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15843015#comment-15843015
]
ASF GitHub Bot commented on FLINK-5667:
---------------------------------------
GitHub user tillrohrmann opened a pull request:
https://github.com/apache/flink/pull/3227
[backport] [FLINK-5667] [state] Synchronize asynchronous checkpointing and
close operation
Backport of #3226 onto `release-1.2` branch.
This PR synchronizes asynchronous checkpointing and close operations of a
StreamTask.
The synchronization prevents that an acknowledged checkpoint gets discarded
and that
a discarded checkpoint gets acknowledged. It achieves this by introducing
an atomic
state variable which guards against late close and acknowledge operations.
cc @uce, @StefanRRichter
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/tillrohrmann/flink
asyncCheckpointingFixBackport
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/flink/pull/3227.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 #3227
----
commit d0a2499f10060f159ad92500338125850a82c0c4
Author: Till Rohrmann <[email protected]>
Date: 2017-01-27T15:26:22Z
[FLINK-5667] [state] Synchronize asynchronous checkpointing and close
operation
This PR synchronizes asynchronous checkpointing and close operations of a
StreamTask.
The synchronization prevents that an acknowledged checkpoint gets discarded
and that
a discarded checkpoint gets acknowledged. It achieves this by introducing
an atomic
state variable which guards against late close and acknowledge operations.
----
> Possible state data loss when task fails while checkpointing
> ------------------------------------------------------------
>
> Key: FLINK-5667
> URL: https://issues.apache.org/jira/browse/FLINK-5667
> Project: Flink
> Issue Type: Bug
> Components: State Backends, Checkpointing
> Affects Versions: 1.2.0, 1.3.0
> Reporter: Till Rohrmann
> Assignee: Till Rohrmann
> Priority: Blocker
> Fix For: 1.2.0, 1.3.0
>
>
> It is possible that Flink loses state data when a {{Task}} fails while a
> checkpoint is being drawn. The scenario is the following:
> Flink has finished the synchronous checkpointing part and starts the
> asynchronous part by creating and submitting a {{AsyncCheckpointRunnable}} to
> an {{Executor}}. This runnable is also registered at the closeable registry.
> If the {{Task}} now fails before the {{AsyncCheckpointRunnable}} has
> completed, it will be closed due to being registered in the closeable
> registry. The closing operation will discard all state handles and cancel all
> runnable state futures. However, it will not stop the runnable from sending
> an acknowledge message to the {{CheckpointCoordinator}}.
> If this message completes the pending checkpoint, then this checkpoint will
> be transformed into a {{CompletedCheckpoint}} which is faulty (some of the
> data has already been deleted). Depending on Flink's configuration, this will
> discard older completed checkpoints and thus we will have state data loss.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
