[jira] [Commented] (FLINK-9983) Savepoints should count as checkpoints when recovering

["Aljoscha Krettek (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/FLINK-9983?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16561619#comment-16561619 ] 

Aljoscha Krettek commented on FLINK-9983:
-----------------------------------------

See FLINK-6328 for reference, where this behaviour was changed. We have to be careful not to count savepoints as retained checkpoints.

> Savepoints should count as checkpoints when recovering
> ------------------------------------------------------
>
>                 Key: FLINK-9983
>                 URL: https://issues.apache.org/jira/browse/FLINK-9983
>             Project: Flink
>          Issue Type: Bug
>          Components: State Backends, Checkpointing
>            Reporter: Aljoscha Krettek
>            Priority: Blocker
>             Fix For: 1.6.0
>
>
> If they are not used when recovering you can get intro problems with duplicate output data when a failure occurs after a savepoint was taken but before the next checkpoint occurs.
> The fix, in the long run, is to differentiate between savepoints that have side effects and those that don't. The former would be used for a "stop-with-savepoint" scenario while the latter is for in-between savepoints. This is harder to implement, so I vote for the easy fix described for now for fixing the duplication problems.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)