[
https://issues.apache.org/jira/browse/HBASE-26073?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Aman Poonia updated HBASE-26073:
--------------------------------
Description:
we have a table on a cluster with 100 regions with default split policy
(SteppingSplitPolicy). This is a small table and will not get loaded with too
much data. Now if region size of table is smaller than the normalizer target
region size than there are chances that normalizer will consider the regions
for merges. But since the number of regions are small split policy will trigger
split on next flush. This is a continuous loop and our cluster will be busy in
these two actions.
We plan to consider number of regions and number of regionservers in creating
plans for normalizer.
was:
we have a table on a cluster with 100 regions with default split policy
(SteppingSplitPolicy). This is a small table and will not get loaded with too
much data. Now if region size of table is smaller than the normalizer target
region size than there are chances that normalizer will consider the regions
for merges. But since the number of regions are small split policy will trigger
split on next flush. This is a continuous loop and our cluster will be busy in
these two actions.
We plan to consider number of regions and number of regions in creating plans
for normalizer.
> Avoid merging regions if tables has not reached the state where regions are
> not split because of number of regions
> ------------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-26073
> URL: https://issues.apache.org/jira/browse/HBASE-26073
> Project: HBase
> Issue Type: Improvement
> Components: Normalizer
> Affects Versions: 1.7.0, 2.4.4
> Reporter: Aman Poonia
> Assignee: Aman Poonia
> Priority: Minor
>
> we have a table on a cluster with 100 regions with default split policy
> (SteppingSplitPolicy). This is a small table and will not get loaded with too
> much data. Now if region size of table is smaller than the normalizer target
> region size than there are chances that normalizer will consider the regions
> for merges. But since the number of regions are small split policy will
> trigger split on next flush. This is a continuous loop and our cluster will
> be busy in these two actions.
> We plan to consider number of regions and number of regionservers in creating
> plans for normalizer.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
