[ 
https://issues.apache.org/jira/browse/HBASE-16659?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

binlijin updated HBASE-16659:
-----------------------------
    Description: 
CellUtil.createFirstOnRow is more effective than KeyValueUtil.createFirstOnRow, 
because there is no byte[] copy.

Current in ReversedRegionScannerImpl#nextRow, to get the previous row to seek, 
it use KeyValueUtil.createFirstOnRow which will have two copy, the first copy 
is copy to a tmp byte[], the second copy is in KeyValue. We can use 
CellUtil.createFirstOnRow which will have no copy.

  was:Current in ReversedRegionScannerImpl#nextRow, to get the previous row to 
seek, it use KeyValueUtil.createFirstOnRow which will have two copy, the first 
copy is copy to a tmp byte[], the second copy is in KeyValue. We can use 
CellUtil.createFirstOnRow which will have no copy.


> Use CellUtil.createFirstOnRow instead of KeyValueUtil.createFirstOnRow in 
> some places. 
> ---------------------------------------------------------------------------------------
>
>                 Key: HBASE-16659
>                 URL: https://issues.apache.org/jira/browse/HBASE-16659
>             Project: HBase
>          Issue Type: Improvement
>            Reporter: binlijin
>            Assignee: binlijin
>            Priority: Minor
>         Attachments: HBASE-16659-master.patch, HBASE-16659-master_v2.patch
>
>
> CellUtil.createFirstOnRow is more effective than 
> KeyValueUtil.createFirstOnRow, because there is no byte[] copy.
> Current in ReversedRegionScannerImpl#nextRow, to get the previous row to 
> seek, it use KeyValueUtil.createFirstOnRow which will have two copy, the 
> first copy is copy to a tmp byte[], the second copy is in KeyValue. We can 
> use CellUtil.createFirstOnRow which will have no copy.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to