[
https://issues.apache.org/jira/browse/FLINK-10354?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16625914#comment-16625914
]
ASF GitHub Bot commented on FLINK-10354:
----------------------------------------
aljoscha commented on issue #6704: [FLINK-10354] Revert "[FLINK-6328] [chkPts]
Don't add savepoints to CompletedCheckpointStore"
URL: https://github.com/apache/flink/pull/6704#issuecomment-423998830
I think falling back to earlier checkpoints/savepoints is orthogonal to this
and covered in https://issues.apache.org/jira/browse/FLINK-4815.
As it is currently, you can have the (quite possible) scenario where you do
a savepoint and then your job fails. This leads to data duplication in the sink.
I think the scenario where you do a savepoint, then delete that savepoint,
and then try to recover is unlikely. And even if it does happen you can
manually fix the situation by editing ZooKeeper entries (I think). There is no
possible manual recovery for the above scenario (downstream systems possible
have already consumed the emitted and committed data).
What do you think?
@tillrohrmann Is this "revert" still valid or where there changes in the
meantime that could break things?
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
> Savepoints should be counted as retained checkpoints
> ----------------------------------------------------
>
> Key: FLINK-10354
> URL: https://issues.apache.org/jira/browse/FLINK-10354
> Project: Flink
> Issue Type: Bug
> Components: State Backends, Checkpointing
> Affects Versions: 1.6.0
> Reporter: Dawid Wysakowicz
> Assignee: Dawid Wysakowicz
> Priority: Major
> Labels: pull-request-available
> Fix For: 1.7.0
>
>
> This task is about reverting [FLINK-6328].
> The problem is that you can get incorrect results with exactly-once sinks if
> there is a failure after taking a savepoint but before taking the next
> checkpoint because the savepoint will also have manifested side effects to
> the sink.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
