[
https://issues.apache.org/jira/browse/PHOENIX-3970?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16058879#comment-16058879
]
James Taylor commented on PHOENIX-3970:
---------------------------------------
I think we need a way to repro the deadlock issue before making a change like
this. Vincent's bug fix for the sorting plus using a high priority queue will
fix deadlock issues IMHO. The n^2 sorting was done while the row is locked
which explains the deadlocks. I don't see how this patch helps.
Samarth - how about doing the partial rebuild triggered from
UngroupedAggregateRegionObserver asynchronously as we do with stats collection?
The client thread in MetaDataRegionObserver could just poll until the async
work is finished. We'd need to do a bit more tracking for each scan being
finished on the index header row.
Not sure all of this is actually necessary with Vincent's sort fix. The intent
is that the partial index rebuild is fast. An alternative would be to disable
index and clear the index_disable_timestamp if the scan times out (since this
is an indication something is wrong).
> Ensure that automatic partial index rebuilds are served from the index
> handler pool
> -----------------------------------------------------------------------------------
>
> Key: PHOENIX-3970
> URL: https://issues.apache.org/jira/browse/PHOENIX-3970
> Project: Phoenix
> Issue Type: Bug
> Reporter: Lars Hofhansl
> Assignee: Lars Hofhansl
> Attachments: 3970.txt, 3970-v2.txt
>
>
> This (and other issues) have rendered multiple larger cluster inoperable.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
