[
https://issues.apache.org/jira/browse/HBASE-13412?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14482280#comment-14482280
]
Lars Hofhansl commented on HBASE-13412:
---------------------------------------
[~stack] Not a fan of ThreadLocals. These tend to hang around. You concerned
about the overhead of creating one?
[~eclark] Is the logic correct?
{code}
+ double sizeJitter = conf.getDouble("hbase.hregion.max.filesize.jitter",
0.02D);
+ this.desiredMaxFileSize = (long)(desiredMaxFileSize * (RANDOM.nextFloat()
- 0.5D) * sizeJitter);
{code}
You mean:
{code}
+ double sizeJitter = conf.getDouble("hbase.hregion.max.filesize.jitter",
0.02D);
+ this.desiredMaxFileSize += (long)(desiredMaxFileSize * (RANDOM.nextFloat()
- 0.5D) * sizeJitter);
{code}
(note the +=)
> Region split decisions should have jitter
> -----------------------------------------
>
> Key: HBASE-13412
> URL: https://issues.apache.org/jira/browse/HBASE-13412
> Project: HBase
> Issue Type: New Feature
> Components: regionserver
> Affects Versions: 1.0.0, 2.0.0
> Reporter: Elliott Clark
> Assignee: Elliott Clark
> Fix For: 2.0.0, 1.1.0
>
> Attachments: HBASE-13412.patch
>
>
> Whenever a region splits it causes lots of IO (compactions are queued for a
> while). Because of this it's important to make sure that well distributed
> tables don't have all of their regions split at exactly the same time.
> This is basically the same as our compaction jitter.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
