[
https://issues.apache.org/jira/browse/FLINK-5626?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15837387#comment-15837387
]
ASF GitHub Bot commented on FLINK-5626:
---------------------------------------
Github user aljoscha commented on the issue:
https://github.com/apache/flink/pull/3198
I think we can keep from having to close the `AbstractStateBackend` by not
caching the `dbOptions` and `columnOptions` and instead giving new ones to the
`RocksDBKeyedStateBackend`. The keyed backend is not meant to be stateless and
already has a lifecycle and will be disposed. It would simply take ownership of
these options as well which would make responsibilities more clear.
Sorry for being annoying about this but I would really like to keep the
"state backend factory" as a factory and not add lifecycle stuff to it as well.
> Improve resource release in RocksDBKeyedStateBackend
> ----------------------------------------------------
>
> Key: FLINK-5626
> URL: https://issues.apache.org/jira/browse/FLINK-5626
> Project: Flink
> Issue Type: Bug
> Components: State Backends, Checkpointing
> Reporter: Stefan Richter
> Assignee: Stefan Richter
>
> I found some minor cases in which some resources in
> `RocksDBKeyedStateBackend` might not get properly released in case of an
> exception, because their ownership is not completely transferred to the
> closing entity.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
