hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Lars Hofhansl (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-13269) Remove result array preallocation to avoid OOME with large scan caching values
Date Wed, 18 Mar 2015 05:26:39 GMT

    [ https://issues.apache.org/jira/browse/HBASE-13269?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14366663#comment-14366663
] 

Lars Hofhansl commented on HBASE-13269:
---------------------------------------

Can't we at least to a Math.min(1000, rows) or something?
Since <rows> is now a limit, there will be few clients that know ahead of time that
they'll want to see only 1000 rows (now that the chunk size is defined by the size limit).
I.e. I'm guessing that rows will likely be Integer.MAX_VALUE or less than 1000.

Or maybe 100 is even better.


> Remove result array preallocation to avoid OOME with large scan caching values
> ------------------------------------------------------------------------------
>
>                 Key: HBASE-13269
>                 URL: https://issues.apache.org/jira/browse/HBASE-13269
>             Project: HBase
>          Issue Type: Bug
>            Reporter: Andrew Purtell
>            Assignee: Andrew Purtell
>             Fix For: 1.0.1, 0.98.12
>
>         Attachments: HBASE-13269-0.98.patch, HBASE-13269-1.0.patch
>
>
> Scan#setCaching(Integer.MAX_VALUE) will likely terminate the regionserver with an OOME
due to preallocation of the result array according to this parameter.  We should limit the
preallocation to some sane value. Definitely affects 0.98 (fix needed to HRegionServer) and
1.0.x (fix needed to RsRPCServices), not sure about later versions. 



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

Mime
View raw message