[
https://issues.apache.org/jira/browse/PHOENIX-4977?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16661493#comment-16661493
]
Karan Mehta commented on PHOENIX-4977:
--------------------------------------
[~vincentpoon]
The property can be set per table however the RegionServer can still be killed
if the property is not set for some other table whose regions are also hosted
by same RegionServer (also assuming that both are phoenix tables with indexes).
It can be confusing to debug such a scenario. Lets simplify this by just having
an RS level global property. Thoughts?
> Make KillServerOnFailurePolicy a configurable option in
> PhoenixIndexFailurePolicy
> ---------------------------------------------------------------------------------
>
> Key: PHOENIX-4977
> URL: https://issues.apache.org/jira/browse/PHOENIX-4977
> Project: Phoenix
> Issue Type: Bug
> Affects Versions: 4.14.0
> Reporter: Vincent Poon
> Assignee: Vincent Poon
> Priority: Major
> Attachments: PHOENIX-4977.v1.4.x-HBase-1.4.patch
>
>
> Currently PhoenixIndexFailurePolicy, which is the default policy, delegates
> to KillServerOnFailurePolicy. This is hardcoded in the constructor.
> Apparently this was added for a specific use case,
> BLOCK_DATA_TABLE_WRITES_ON_WRITE_FAILURE, and the policy itself derives from
> the days where forcing a RS kill was in effect the way to 'rebuild' the index
> via WAL replay.
> There are still cases where it's applicable, such as when Syscat itself
> cannot be updated in order to e.g. disable an index. However, killing the RS
> may be too aggressive for some, who might prefer a temporarily out of sync
> index to a potentially cascading wave of aborts.
> We should add a config option to control this.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
