[
https://issues.apache.org/jira/browse/HBASE-15154?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15111915#comment-15111915
]
stack commented on HBASE-15154:
-------------------------------
bq. When master host META, we dont want a block cache (small size) so that META
table block will get inside that?
This seems to be happening in hbase 1.1, even when we have master not carry
regions (default)
> Master puts up a blockcache instance
> ------------------------------------
>
> Key: HBASE-15154
> URL: https://issues.apache.org/jira/browse/HBASE-15154
> Project: HBase
> Issue Type: Bug
> Reporter: stack
> Priority: Critical
>
> Master is putting up a blockcache instance. It shouldn't. This issue comes of
> the unification of Master and RegionServer where Master is a subclass of
> RegionServer. Our [~jmspaggi] found out the hard way today when he tried to
> bring up a cluster with a large offheap cache only the master member had been
> given less memory.
> Marking critical rather than blocker because there is a means of configuring
> your way out of this little entanglement; set hbase.bucketcache.size to zero
> in the master config only.
> If you want to set direct memory for regionservers only and not on the
> master, you should do the following:
> Set HBASE_OFFHEAPSIZE=0G
> ... and then turn it on for RegionServers only:
> HBASE_REGIONSERVER_OPTS="HBASE_REGIONSERVER_OPTS
> -XX:MaxDirectMemorySize=XXXG" where XXX is size of the direct memory you want
> to allocate.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
