[
https://issues.apache.org/jira/browse/FLINK-7731?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16200409#comment-16200409
]
Aljoscha Krettek commented on FLINK-7731:
-----------------------------------------
Ah yes, sorry for not noticing that earlier. The combination of global windows
(which don't have a good cleanup time), trigger firings and empty windows is
indeed complex. There is also FLINK-5363 which aims at chaining the semantics a
bit but it is highly likely that this introduces yet more different problems.
I think for now the behaviour is OK but we might need to improve documentation
(in Javadocs) a bit. What do you think?
> Trigger on GlobalWindow does not clean state completely
> -------------------------------------------------------
>
> Key: FLINK-7731
> URL: https://issues.apache.org/jira/browse/FLINK-7731
> Project: Flink
> Issue Type: Bug
> Components: Core, DataStream API
> Affects Versions: 1.3.2
> Reporter: Gerard Garcia
> Priority: Minor
>
> I have an operator that consists of:
> CoGroup Datastream -> GlobalWindow -> CustomTrigger -> Apply function
> The custom trigger fires and purges the elements after it has received the
> expected number of elements (or when a timeout fires) from one of the streams
> and the apply function merges the elements with the ones received from the
> other stream. It appears that the state of the operator grows continuously so
> it seems it never gets completely cleaned.
> There is a discussion in
> http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/Clean-GlobalWidnow-state-td15613.html
> that suggests that it may be a bug.
> This job reproduces the issue:
> https://github.com/GerardGarcia/flink-global-window-growing-state
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
