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-10660) MR over snapshots can OOM when alternative blockcache is enabled
Date Tue, 04 Mar 2014 01:32:21 GMT

    [ https://issues.apache.org/jira/browse/HBASE-10660?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13918897#comment-13918897
] 

Lars Hofhansl commented on HBASE-10660:
---------------------------------------

We should just disable the caching of data blocks (i.e. just setting cacheBlocks to false
on the Scan object used). We still want the index blocks cached (typically a few MB are enough
for those) otherwise scan performance will probably be quite bad.


> MR over snapshots can OOM when alternative blockcache is enabled
> ----------------------------------------------------------------
>
>                 Key: HBASE-10660
>                 URL: https://issues.apache.org/jira/browse/HBASE-10660
>             Project: HBase
>          Issue Type: Bug
>          Components: mapreduce
>    Affects Versions: 0.98.0, 0.99.0
>            Reporter: Nick Dimiduk
>            Assignee: Nick Dimiduk
>         Attachments: HBASE-10660.00.patch
>
>
> Running {{IntegrationTestTableSnapshotInputFormat}} with the {{BucketCache}} enabled
results in OOM. The map task is running a sequential scan over the region region it opened,
so probably it's not benefiting much from a blockcache. Just disable blockcache entirely for
these scans because it's likely the cache config detected is designed for a RS running on
a different class of hardware than that running the map task.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Mime
View raw message