[
https://issues.apache.org/jira/browse/PHOENIX-3995?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16086359#comment-16086359
]
James Taylor commented on PHOENIX-3995:
---------------------------------------
To support this, I believe now that we've got a reasonable way of isolating the
config used for UPSERT SELECT with PHOENIX-3994, all that is required is to
- update PhoenixRpcScheduler with a new queue for UPSERT SELECT at a priority
between normal and the index priority.
- create a new UpsertSelectRpcControllerFactory factory that uses the new
priority
- set the controller factory in the upsertSelectConfig we've setup in
UngroupedAggregateRegionObserver
Does that cover it, [~samarthjain]?
> Use medium priority for UPSERT SELECT distributed across cluster
> ----------------------------------------------------------------
>
> Key: PHOENIX-3995
> URL: https://issues.apache.org/jira/browse/PHOENIX-3995
> Project: Phoenix
> Issue Type: Bug
> Reporter: James Taylor
> Assignee: Samarth Jain
>
> An UPSERT SELECT that is running on the server-side should use a medium
> priority (lower than index priority but higher than normal priority) to
> prevent deadlocks. If we use the index priority, then there's a danger of
> deadlocks occurring if the table being upserted into also as indexes (as the
> index updates would be processed at the index priority).
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
