[
https://issues.apache.org/jira/browse/FLINK-22049?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17311481#comment-17311481
]
Yun Tang commented on FLINK-22049:
----------------------------------
I agree that getting start key group could be simplified, [~kkl0u] why you
implement current case in FLINK-4892 ?
> Simplify calculating the starting index of the local key-group range in
> InternalTimerServiceImpl constructor.
> -------------------------------------------------------------------------------------------------------------
>
> Key: FLINK-22049
> URL: https://issues.apache.org/jira/browse/FLINK-22049
> Project: Flink
> Issue Type: Improvement
> Components: API / DataStream
> Reporter: ZhangWei
> Priority: Minor
>
> For org.apache.flink.streaming.api.operators.InternalTimerServiceImpl;
> in current implementation, when we try to find the starting index of the
> local key-group range, we iterate over the KeyGroupRange and try to find the
> min value.
> while this is unnecessary and wasteful, because the KeyGroupRange is often
> monotonically increasing and we can just get the startKeyGroup which is what
> we want.
> It is even worse when we set a very large max parallelism(eg: 10k level) and
> a small parallelism(eg: 10 level), thus a KeyGroupRange may contain about 1k
> index in it. And then we iterate this 1k increasing numbers to find the min
> value.
> so if the KeyGroupRange is not an EMPTY_KEY_GROUP_RANGE(i.e.
> KeyGroupRange#getNumberOfKeyGroups() > 0), we can just get the startKeyGroup.
> if KeyGroupRange is just an EMPTY_KEY_GROUP_RANGE, use the Integer.MAX_VALUE.
> I think this can avoid much unnecessary operation and save time.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
