[
https://issues.apache.org/jira/browse/FLINK-11188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16727251#comment-16727251
]
Hequn Cheng commented on FLINK-11188:
-------------------------------------
[~fhueske] Hi, thanks for your reply. I think you are right. Rows-bounded over
should support retention time. The state will not be cleaned up automatically.
The accumulators in rows-bounded over will be kept forever similar to the
unbounded group by.
I find that range-bounded over can't automatically clean up their state either.
Under the current implementation, accumulators are only cleaned by retention
time strategy. We could:
- add automatic cleanup logic for them.
- stick to retention time strategy.
I personally prefer the first option.
- The dedicate cleanup logic can both support rowtime and proctime.
- It can clean up the state automatically, i.e., users don't need to know it.
- It only removes expired data according to the range of over window which will
not result in incomplete/invalid results
What do you think?
> Bounded over should not enable state retention time
> ----------------------------------------------------
>
> Key: FLINK-11188
> URL: https://issues.apache.org/jira/browse/FLINK-11188
> Project: Flink
> Issue Type: Bug
> Components: Table API & SQL
> Reporter: Hequn Cheng
> Assignee: Hequn Cheng
> Priority: Major
>
> As discussed in FLINK-11172, time-based operations (GROUP BY windows, OVER
> windows, time-windowed join, etc.) are inherently bound by time and
> automatically clean up their state. We should not add state cleanup or TTL
> for these operators.
> If I understand correctly, we should not add the retention logic for
> rows-bounded operations either. I think we should disable state retention
> logic for:
> - ProcTimeBoundedRangeOver
> - ProcTimeBoundedRowsOver
> - RowTimeBoundedRangeOver
> - RowTimeBoundedRowsOver
> Any suggestions are appreciated!
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
