hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "linbao111 (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-14611) Backport HBASE-14473 (Compute region locality in parallel) to 0.98
Date Wed, 23 Dec 2015 12:48:46 GMT

    [ https://issues.apache.org/jira/browse/HBASE-14611?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15069591#comment-15069591
] 

linbao111 commented on HBASE-14611:
-----------------------------------

more to be added,i think hbase.master.balancer.regionLocationCacheTime default value after
patch did a serious impact on hdfs
namenode(hdfs-audit.log become huge,and full of hmaster accessing)

so if any one update to hbase 0.98.16.1,rememer to add  hbase.master.balancer.regionLocationCacheTime
to be 30*2*60 in hbase-site.xml,

> Backport HBASE-14473 (Compute region locality in parallel) to 0.98
> ------------------------------------------------------------------
>
>                 Key: HBASE-14611
>                 URL: https://issues.apache.org/jira/browse/HBASE-14611
>             Project: HBase
>          Issue Type: Task
>            Reporter: Andrew Purtell
>            Assignee: Andrew Purtell
>             Fix For: 0.98.16
>
>         Attachments: HBASE-14611-0.98.patch
>
>
> [~eclark] contributed a nice change on HBASE-14473 that scales calculation of locality
balance cost to larger clusters. I'd like to bring this back to 0.98 for folks running it
on larger clusters. The changes require a partial rewrite of RegionLocationFinder hence a
backport issue.
> The difference between this backport and RegionLocationFinder on later branches is we
preserve the ability to change the expiration time of cache items with the configuration parameter
"hbase.master.balancer.regionLocationCacheTime". 
> The difference between RegionLocationFinder in 0.98 before and after this change is the
expiration time of cache entries will be set according to when they are written into the cache
instead of from when they are last accessed, which seems better to me.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message