[
https://issues.apache.org/jira/browse/HBASE-24664?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17149092#comment-17149092
]
Duo Zhang commented on HBASE-24664:
-----------------------------------
It is the old behavior for a long time. So the question here should not be 'I'm
not sure if there is a scenario that the old one performs better', it should be
'I'm not sure if the new one always performs better.'
We should have a way to let users just keep the old behavior. Changing this may
lead to a split storm on an existing cluster...
> Some changing of split region by overall region size rather than only one
> store size
> ------------------------------------------------------------------------------------
>
> Key: HBASE-24664
> URL: https://issues.apache.org/jira/browse/HBASE-24664
> Project: HBase
> Issue Type: Improvement
> Components: regionserver
> Reporter: Zheng Wang
> Assignee: Zheng Wang
> Priority: Major
>
> As a distributed cluster, HBase distribute loads in unit of region, so if
> region grows too big,
> it will bring some negative effects, such as:
> 1. Harder to homogenize disk usage(consider locality)
> 2. Might cost more time on region opening
> 3. After split, the daughter region might lead to more io cost on compaction
> in a short time(if write evenly)
> HBASE-24530 introduced a new SteppingAllStoresSizeSplitPolicy, and as
> discussed in its comments and related
> [thread|https://lists.apache.org/thread.html/r08a8103e2532eb667a0fcb4efa8a4117b3f82e6251bc4bd0bc157c26%40%3Cdev.hbase.apache.org%3E],
> we should do follow-on tasks in this new issue.
> 1. Set SteppingAllStoresSizeSplitPolicy as default
> 2. Mark SteppingSplitPolicy and IncreasingToUpperBoundRegionSplitPolicy as
> deprecated
> 3. Fix ConstantSizeRegionSplitPolicy to split region by overall region size
> also
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
