[ 
https://issues.apache.org/jira/browse/HBASE-25281?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

niuyulin updated HBASE-25281:
-----------------------------
    Description: 
[https://github.com/apache/hbase/blob/master/hbase-server/src/main/java/org/apache/hadoop/hbase/tool/BulkLoadHFilesTool.java#L688]

if hfile span multi regions, for example A,B,C,D,E,F(the start key of these 
regions are in ascending order), we should use region C endkey to split, not 
region A. In this way, we can get equal .top and .bottom hfiles, reduce time 
complexity of split from O(n) to O(logn),decrease invoke of {{bulkLoad}} rpc to 
regionserver ,and also avoid write amplification during {{copyHFileHalf}}

  was:
[https://github.com/apache/hbase/blob/master/hbase-server/src/main/java/org/apache/hadoop/hbase/tool/BulkLoadHFilesTool.java#L688]

if hfile span multi regions, for example A,B,C,D,E,F(the start key of these 
regions are in ascending order), we should use region C endkey to split, not 
region A. In this way, we can get equal .top and .bottom hfiles, reduce time 
complexity of split from O(n) to O(logn),decrease invoke of {{bulkLoad}} rpc to 
regionserver ,and also avoid write amplification during {{copyHFileHalf}}


> Bulkload split hfile too many times due to unreasonable split point
> -------------------------------------------------------------------
>
>                 Key: HBASE-25281
>                 URL: https://issues.apache.org/jira/browse/HBASE-25281
>             Project: HBase
>          Issue Type: Improvement
>          Components: tooling
>            Reporter: niuyulin
>            Assignee: niuyulin
>            Priority: Minor
>         Attachments: bulkload.png
>
>
> [https://github.com/apache/hbase/blob/master/hbase-server/src/main/java/org/apache/hadoop/hbase/tool/BulkLoadHFilesTool.java#L688]
> if hfile span multi regions, for example A,B,C,D,E,F(the start key of these 
> regions are in ascending order), we should use region C endkey to split, not 
> region A. In this way, we can get equal .top and .bottom hfiles, reduce time 
> complexity of split from O(n) to O(logn),decrease invoke of {{bulkLoad}} rpc 
> to regionserver ,and also avoid write amplification during {{copyHFileHalf}}



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to