[
https://issues.apache.org/jira/browse/FLINK-17058?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17083045#comment-17083045
]
Aljoscha Krettek commented on FLINK-17058:
------------------------------------------
Sorry, we had public vacation on Friday and Monday in Germany... 😅
Thanks for the code! I think there are some very tricky corner cases or things
that we learned over time that could be problematic:
- the namespace of state is not changed when calling the nested trigger, i.e.
if the nested trigger sets state or timers this could interfere with the
timeout trigger because we just straight pass through the context
- {{onProcessingTime()}} doesn't invoke the nested trigger, which might want
to do a purge or other things
- same for {{onEventTime()}} where this wrapping trigger would not
react/forward at all
I don't know if we can find a good solution that would be useful for many Flink
users with the current restrictions of the APIs. By the way, there was also
FLINK-4407 which we eventually dropped because it was considered too much
effort for the potential benefit.
> Adding TimeoutTrigger support nested triggers
> ---------------------------------------------
>
> Key: FLINK-17058
> URL: https://issues.apache.org/jira/browse/FLINK-17058
> Project: Flink
> Issue Type: Improvement
> Components: API / DataStream
> Reporter: Roey Shem Tov
> Priority: Minor
> Attachments: ProcessingTimeoutTrigger.java
>
>
> Hello,
> first Jira ticket that im opening here so if there is any mistakes of how
> doing it, please recorrect me.
> My suggestion is to add a TimeoutTrigger that apply as nestedTrigger(as
> example how the PurgeTrigger does).
> The TimeoutTrigger will support ProcessTime/EventTime and will have 2 timeous:
> # Const timeout - when the first element of the window is arriving it is
> opening a timeout of X millis - after that the window will be evaluate.
> # Continual timeout - each record arriving will increase the timeout of the
> evaluation of the window.
> Â
> I found it very useful in our case when using flink, and i would like to work
> on it (if it is possible).
> what do you think?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
