[jira] [Commented] (HBASE-7347) Allow multiple readers per storefile

["Lars Hofhansl (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HBASE-7347?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13531364#comment-13531364 ] 

Lars Hofhansl commented on HBASE-7347:
--------------------------------------

Options might include:
* Ad hoc fix in HBASE-7336 is good enough (using seek + read when possible without contention, false back to p-read when needed)
* A separate reader for compactions (Stack suggested that)
* A reader for each scanner that has cacheBlocks set to false (presumably because it is expected to be large scan, that would also cover compactions)
* multiple readers for store files, round robin between them
** based on size (i.e. a reader per each n GB)
** based on access pattern (i.e. create readers, cache them for a bit, expire them)

What probably does not work:
* Use p-read everywhere (does not perform well)
* A reader per scanner (probably not efficient, open is a name node operation)


                
> Allow multiple readers per storefile
> ------------------------------------
>
>                 Key: HBASE-7347
>                 URL: https://issues.apache.org/jira/browse/HBASE-7347
>             Project: HBase
>          Issue Type: Bug
>            Reporter: Lars Hofhansl
>
> Currently each store file is read only through the single reader regardless of how many concurrent read requests access that file.
> This issue is to explore alternate designs.

--
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