[
https://issues.apache.org/jira/browse/HUDI-3661?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17630751#comment-17630751
]
Danny Chen commented on HUDI-3661:
----------------------------------
Thanks, maybe you are right, let's see if there is back pressure when removing
the async executor.
> flink async compaction is not threadsafe when use watermark
> -----------------------------------------------------------
>
> Key: HUDI-3661
> URL: https://issues.apache.org/jira/browse/HUDI-3661
> Project: Apache Hudi
> Issue Type: Bug
> Reporter: hd zhou
> Priority: Major
> Attachments: image-2022-03-18-19-38-39-257.png
>
>
> async compaction will start a executor async compaciton and send compaction
> result message to next flink operator. But collector.collect() is not a
> threadsafe function. when use watermark or latencyMarker, they both call
> collector.collect() may cause issue.
> we should not let async compaction = false
>
> !image-2022-03-18-19-38-39-257.png!
>
>
> !https://git.bilibili.co/datacenter/bili-hudi/uploads/79608d01b0301de84d1d9e3cf24f1d21/image.png!
>
> !https://git.bilibili.co/datacenter/bili-hudi/uploads/e9c2f27d395e708a407bcf40f672c870/image.png!
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
