[
https://issues.apache.org/jira/browse/PHOENIX-5945?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17144468#comment-17144468
]
Abhishek Singh Chouhan commented on PHOENIX-5945:
-------------------------------------------------
Planning to repurpose this to change the split policy of the table so that it
doesn't split for a near term solution [~ckulkarni]
> TaskRegionObserver can kick off the same task multiple times if SYSTEM.TASK
> has split
> -------------------------------------------------------------------------------------
>
> Key: PHOENIX-5945
> URL: https://issues.apache.org/jira/browse/PHOENIX-5945
> Project: Phoenix
> Issue Type: Bug
> Affects Versions: 4.15.0
> Reporter: Chinmay Kulkarni
> Priority: Major
> Fix For: 5.1.0, 4.16.0
>
>
> We don't specify a split policy for
> [SYSTEM.TASK|https://github.com/apache/phoenix/blob/5f9364db7e4925229704706e148e62f4cf4ec4c2/phoenix-core/src/main/java/org/apache/phoenix/query/QueryConstants.java#L381],
> so by default it will be allowed to split. Now if SYSTEM.TASK spans multiple
> regions, each region's
> [postOpen|https://github.com/apache/phoenix/blob/5f9364db7e4925229704706e148e62f4cf4ec4c2/phoenix-core/src/main/java/org/apache/phoenix/coprocessor/TaskRegionObserver.java#L137]
> schedules the SelfHealingTask at the specified interval and so [each region
> will run a FTS on the
> table|https://github.com/apache/phoenix/blob/5f9364db7e4925229704706e148e62f4cf4ec4c2/phoenix-core/src/main/java/org/apache/phoenix/coprocessor/TaskRegionObserver.java#L159]
> and try to kick-off all the incomplete and non-failed tasks.
> This can lead to the same tasks being kicked off multiple times as a corner
> race condition in spite of [this
> check|https://github.com/apache/phoenix/blob/5f9364db7e4925229704706e148e62f4cf4ec4c2/phoenix-core/src/main/java/org/apache/phoenix/coprocessor/TaskRegionObserver.java#L187-L195]
> (which is another FTS) and also lead to unnecessary extra load on the server.
> We do not explicitly outline that tasks need to be idempotent, so we should
> handle this properly in the TaskRegionObserver so that each region is only
> responsible for tasks lying within its boundaries.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
