[
https://issues.apache.org/jira/browse/FLINK-37107?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Zakelly Lan updated FLINK-37107:
--------------------------------
Description: In FLINK-37089, we resolved the deadlock of derived requests
waiting on draining current request, when user call `asyncProcessWithKey`.
However, we also found the timer may also suffer from this problem. Thus I
suggest a buffer overdraft for all those derived requests, to resolve the
deadlock when there is a single hotspot key and all the requests queue in line
while there comes a derived with same key, that cannot properly queue in a full
buffer. (was: In FLINK-37089, we resolved the deadlock of derived requests
waiting on draining current request, when user call `asyncProcessWithKey`.
However, we also found the timer may also suffer from this problem. Thus I
suggest a buffer overdraft for all those derived requests, to resolve the
deadlock when there is a single hotspot key and all the requests queue in line
while there comes a derived with same key cannot properly queue in a full
buffer.)
> Allow buffer overdraft in async state processing
> ------------------------------------------------
>
> Key: FLINK-37107
> URL: https://issues.apache.org/jira/browse/FLINK-37107
> Project: Flink
> Issue Type: Sub-task
> Reporter: Zakelly Lan
> Assignee: Zakelly Lan
> Priority: Major
>
> In FLINK-37089, we resolved the deadlock of derived requests waiting on
> draining current request, when user call `asyncProcessWithKey`. However, we
> also found the timer may also suffer from this problem. Thus I suggest a
> buffer overdraft for all those derived requests, to resolve the deadlock when
> there is a single hotspot key and all the requests queue in line while there
> comes a derived with same key, that cannot properly queue in a full buffer.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
