[
https://issues.apache.org/jira/browse/FLINK-4169?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Aljoscha Krettek resolved FLINK-4169.
-------------------------------------
Resolution: Fixed
Done in
https://github.com/apache/flink/commit/a710fc3d58014a7191de61fd5d6370de97c9e75e
> CEP Does Not Work with RocksDB StateBackend
> -------------------------------------------
>
> Key: FLINK-4169
> URL: https://issues.apache.org/jira/browse/FLINK-4169
> Project: Flink
> Issue Type: Bug
> Components: CEP
> Affects Versions: 1.0.0, 1.1.0, 1.0.1, 1.0.2, 1.0.3
> Reporter: Aljoscha Krettek
> Assignee: Aljoscha Krettek
> Priority: Blocker
>
> A job will never match any patterns because {{ValueState.update()}} is not
> called in the keyed CEP operators for updating the {{NFA}} state and the
> priority queue state.
> The reason why it works for other state backends is that they are very lax in
> their handling of state: if the object returned from {{ValueState.value())}}
> is mutable changes to this will be reflected in checkpoints even if
> {{ValueState.update()}} is not called. RocksDB, on the other hand, does
> always deserialize/serialize state values when accessing/updating them, so
> changes to the returned object will not be reflected in the state unless
> {{update()}} is called.
> We should fix this and also add a test for it. This might be tricky because
> we have to pull together RocksDB and CEP.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
