[
https://issues.apache.org/jira/browse/FLINK-4754?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15876543#comment-15876543
]
ASF GitHub Bot commented on FLINK-4754:
---------------------------------------
Github user StephanEwen commented on the issue:
https://github.com/apache/flink/pull/3374
Hi @tony810430 thank you for the pull request!
The code looks good.
My feeling is, though, that the number of checkpoints to retain is
something that we want rather in the configuration of the JobManager, than in
the programs snapshot settings.
Think of it like that: There are often two roles, *developer* and *ops*.
- The developer writes the streaming program, and sets the values for the
snapshot settings, like what checkpoint interval would work well for the
application etc.
- The ops person writes the cluster's config and is concerned with
running the job reliably.
Having multiple retained checkpoints is something that concerns more the
ops person.
What do you think?
> Make number of retained checkpoints user configurable
> -----------------------------------------------------
>
> Key: FLINK-4754
> URL: https://issues.apache.org/jira/browse/FLINK-4754
> Project: Flink
> Issue Type: Improvement
> Components: State Backends, Checkpointing
> Affects Versions: 1.1.2
> Reporter: Ufuk Celebi
> Assignee: Wei-Che Wei
>
> The number of retained successful checkpoints is fixed to 1. Expose this
> setting via the {{CheckpointConfig}} instead of having it fixed as a static
> field in the {{CheckpointRecoveryFactory}}.
> With the current state of things, this would require to set this value lazily
> in the checkpoint store implementations instead of setting it when creating
> the store.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
