[ 
https://issues.apache.org/jira/browse/HBASE-27607?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

mokai updated HBASE-27607:
--------------------------
    Description: 
I found some regions in our production cluster have one huge hfile and can not 
split automatically anymore, it causes the poor read performance.

If a region loaded lots of data via bulkload, it will split after compaction 
done. The reference files are cleared by 'CompactedHFilesDischarger' 
asynchronized, so the daughter regions will not split after compaction 
finished. If the daughter regions have no more write operation, they will never 
split although the rewrote hfile exceed the disired  max hfile 
size(ConstantSizeRegionSplitPolicy), 

!image-2023-02-02-11-25-16-499.png!

  was:
I found some regions in our production cluster have one huge hfile and can not 
split automatically anymore, it causes the poor read performance.

If a region loaded lots of data via bulkload, it will split after compaction 
done. The reference files are cleared by 'CompactedHFilesDischarger' 
asynchronized, so the daughter regions will not split after compaction 
finished. If the daughter regions have no more write operation, they will never 
split although the rewrote hfile exceed the disired  max hfile 
size(ConstantSizeRegionSplitPolicy), 


> Region with one huge hfile should split
> ---------------------------------------
>
>                 Key: HBASE-27607
>                 URL: https://issues.apache.org/jira/browse/HBASE-27607
>             Project: HBase
>          Issue Type: Bug
>    Affects Versions: 2.4.14
>            Reporter: mokai
>            Assignee: mokai
>            Priority: Major
>
> I found some regions in our production cluster have one huge hfile and can 
> not split automatically anymore, it causes the poor read performance.
> If a region loaded lots of data via bulkload, it will split after compaction 
> done. The reference files are cleared by 'CompactedHFilesDischarger' 
> asynchronized, so the daughter regions will not split after compaction 
> finished. If the daughter regions have no more write operation, they will 
> never split although the rewrote hfile exceed the disired  max hfile 
> size(ConstantSizeRegionSplitPolicy), 
> !image-2023-02-02-11-25-16-499.png!



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to