[ 
https://issues.apache.org/jira/browse/HBASE-7952?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13589162#comment-13589162
 ] 

Raymond Liu commented on HBASE-7952:
------------------------------------

Hi Stack

I think for the normal code path, the KV should also be sorted between 
different storefiles by the heap before feeding to matcher in storescanner? So 
it will be generated to be in order.

While I am not familiar with the history of this part of code. So I googled 
matcher.update() yesterday try to find out why it is in the way.

And as 
http://people.apache.org/~psmith/hbase/sandbox/hbase/hbase-core/xref/org/apache/hadoop/hbase/regionserver/StoreFileGetScan.html,
 I can see there are some class do scan single storefile without comparing to 
the other storefile in around hbase 0.20 version. 

52     * <p>
53     * This style of StoreFile scanning goes through each
54     * StoreFile in its entirety, most recent first, before
55     * proceeding to the next StoreFile.
56     * <p>

I don't know exact usage for this class, while I think this kind of usage is 
not used anymore? 

                
> Remove update() and Improve ExplicitColumnTracker performance.
> --------------------------------------------------------------
>
>                 Key: HBASE-7952
>                 URL: https://issues.apache.org/jira/browse/HBASE-7952
>             Project: HBase
>          Issue Type: Improvement
>          Components: regionserver
>    Affects Versions: 0.94.1, 0.94.5
>            Reporter: Raymond Liu
>            Assignee: Raymond Liu
>             Fix For: 0.96.0
>
>         Attachments: HBASE_7952.patch
>
>
> In ColumnTracker.java, the update() method is not used by anyone now. And no 
> one will call checkColumn for different HFiles with update() in between files 
> to re-walk through the target columns. All columns will be feed to 
> checkColumn() in order.
> So, within ExplicitColumnTracker, the target columns can be optimized to not 
> dynamic maintain a changing list of columns yet to match. Instead, just move 
> index through it is enough.
> with this optimization to save the time for avoid reconstruct a columns array 
> upon each row, the checkColumn method's performance could be improved by 
> 10-20%.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Reply via email to