[ 
https://issues.apache.org/jira/browse/HBASE-14643?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14963636#comment-14963636
 ] 

ramkrishna.s.vasudevan commented on HBASE-14643:
------------------------------------------------

bq.We should flush the Memstore before get StoreFileWrapper...
Oh!! On checking the close() code it does more stuff including holding the lock 
and then ensuring proper flush and compaction happens. I think directly doing a 
flush wil have some implicaitons?  Because it is only after close() the parent 
actually stops accepting mutations right?

> Avoid Splits from once again opening a closed reader for fetching the first 
> and last key
> ----------------------------------------------------------------------------------------
>
>                 Key: HBASE-14643
>                 URL: https://issues.apache.org/jira/browse/HBASE-14643
>             Project: HBase
>          Issue Type: Improvement
>          Components: regionserver
>    Affects Versions: 2.0.0
>            Reporter: ramkrishna.s.vasudevan
>            Assignee: Heng Chen
>         Attachments: HBASE-14643.patch, HBASE-14643_v1.patch
>
>
> Currently split flow is such that we close the parent region and all its 
> store file readers are also closed.  After that inorder to split the 
> reference files we need the first and last keys for which once again open the 
> readers on those store files.  This could be costlier operation considering 
> the fact that it has to contact the HDFS for this close and open operation. 
> This JIRA is to see if we can improve this.



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

Reply via email to