[
https://issues.apache.org/jira/browse/HBASE-15128?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15120683#comment-15120683
]
Enis Soztutar commented on HBASE-15128:
---------------------------------------
bq. what are all these Tracker that we are adding? and why are we using zk to
store persistent data when we said that zk should be used only for non
transient states and the only exception is replication (and we are trying to
get rid of that)?
I think the patch follows the same model that balancer, catalog janitor and
normalizer already use. That data is still at zookeeper. Agreed that this data
should not be in zk, but it is a different issue.
> Disable region splits and merges in HBCK
> ----------------------------------------
>
> Key: HBASE-15128
> URL: https://issues.apache.org/jira/browse/HBASE-15128
> Project: HBase
> Issue Type: Bug
> Reporter: Enis Soztutar
> Assignee: Heng Chen
> Fix For: 2.0.0, 1.3.0
>
> Attachments: HBASE-15128.patch, HBASE-15128_v1.patch,
> HBASE-15128_v3.patch
>
>
> In large clusters where region splits are frequent, and HBCK runs take
> longer, the concurrent splits cause further problems in HBCK since HBCK
> assumes a static state for the region partition map. We have just seen a case
> where HBCK undo's a concurrently splitting region causing number of
> inconsistencies to go up.
> We can have a mode in master where splits and merges are disabled like the
> balancer and catalog janitor switches. Master will reject the split requests
> if regionservers decide to split. This switch can be turned on / off by the
> admins and also automatically by HBCK while it is running (similar to
> balancer switch being disabled by HBCK).
> HBCK should also disable the Catalog Janitor just in case.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
