[
https://issues.apache.org/jira/browse/FLINK-8968?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16401952#comment-16401952
]
ASF GitHub Bot commented on FLINK-8968:
---------------------------------------
Github user sihuazhou commented on the issue:
https://github.com/apache/flink/pull/5705
@StefanRRichter Sorry for pinging you again in this PR, but I found there
maybe be another issue for `full checkpoint`. It's about the
`kvStateInformation`, we should create a deep copy of `kvStateInformation` when
taking snapshot, currently we use the code `this.kvStateInformationCopy = new
ArrayList<>(stateBackend.kvStateInformation.values());`, but this is not a deep
copy indeed because the `Serializers` in `RegisteredKeyedBackendStateMetaInfo`
maybe not stateless. If you also think this is a problem, instead of make a new
PR for that I'd like to change the code to cover it in this PR, WDYT?
> Fix native resource leak caused by ReadOptions
> -----------------------------------------------
>
> Key: FLINK-8968
> URL: https://issues.apache.org/jira/browse/FLINK-8968
> Project: Flink
> Issue Type: Bug
> Components: State Backends, Checkpointing
> Affects Versions: 1.5.0
> Reporter: Sihua Zhou
> Assignee: Sihua Zhou
> Priority: Major
> Fix For: 1.5.0
>
>
> We should pull the creation of ReadOptions out of the loop in
> {{RocksDBFullSnapshotOperation.writeKVStateMetaData()}}.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
