[
https://issues.apache.org/jira/browse/HBASE-20768?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16531845#comment-16531845
]
Mike Drob commented on HBASE-20768:
-----------------------------------
There is already a comment near the top {code}// Do not scheduleFullRefresh at
master startup{code} so it seems like this might be a regression? related to
AMv2? Patch looks fine to me.
> should we bypass the RegionLocationFinder#scheduleFullRefresh execution when
> master has not initialized
> -------------------------------------------------------------------------------------------------------
>
> Key: HBASE-20768
> URL: https://issues.apache.org/jira/browse/HBASE-20768
> Project: HBase
> Issue Type: Improvement
> Components: Balancer
> Reporter: chenxu
> Priority: Major
> Attachments: HBASE-20768-master-v1.patch
>
>
> Our KYLIN cluster has about 300K Regions, when master startup, It takes a
> long time to do RegionLocationFinder#scheduleFullRefresh.
> During this period, CUBE cannot be built, because master has not initialized.
> Although we can ignore locality factor through HBASE-18478, but i think it
> would be better if we bypass the RegionLocationFinder#scheduleFullRefresh
> until the master initialized
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
