[
https://issues.apache.org/jira/browse/HBASE-25185?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17214344#comment-17214344
]
Michael Stack commented on HBASE-25185:
---------------------------------------
Oh, and regards FN.... let me do a follow-on to deprecate the FN balancer. The
work to integrate FN into stochastic balancer while stalled, seems like good
work. Let me take a looks to see if can kickstart it; i.e. deprecating parts of
FN makes sense but removing all of the FN machinery now seems premature (even
though it likely doesn't work, is not used, and needs finishing).
> Remove the usage of MetaTableAccessor in hbase-balancer
> -------------------------------------------------------
>
> Key: HBASE-25185
> URL: https://issues.apache.org/jira/browse/HBASE-25185
> Project: HBase
> Issue Type: Sub-task
> Components: Balancer, meta
> Reporter: Duo Zhang
> Priority: Major
>
> Let's try to find a way to move MTA to hbase-server. Putting it in
> hbase-balancer is a bit weird (Why is the MetaTableAccessor in the balancer
> module?).
> MetaTableAccessor (MTA) is in hbase-balancer because the favored nodes (FN)
> feature wants to write to meta.
> The FN feature (HBASE-7932 which arrived in hbase-0.96.0) I believe is
> unused. The idea of maintaining locality across a crash is a good one but
> having to enable a different balancer to default to enable FN operation has
> made it so the feature has been let rot. HBASE-21014 makes an appeal for
> integrating FN into the default stochastic balancer. HBASE-15531 is an old
> effort that actually tried to integrate FN into stochastic balancer but
> stalled out.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)