[
https://issues.apache.org/jira/browse/HBASE-723?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12611553#action_12611553
]
Jb Lee commented on HBASE-723:
------------------------------
stack:
I confused with old source file (0.1.3?), where the cachedRegionLocations are
declared as Map<Text, SortedMap<Text, HRegionLocation>>. Sorry.
I agree with you. With soft references, it may not be the problem.
I checked another cache in hbase. HBASE-288 added block cache implementation
in HStoreFile using BlockFSInputStream.
"Map<Long, byte[]> blocks" stores blocks, but I can't find the part that
remove it before the BlockFSInputStream object is GCed.
I think it can be significant problem if we use block cached table because the
block cache is in regionserver and block data is not so small.
Do we need to fix it?
Andrew Purtell:
Hypothetically. Sorry again.
> TableServers's cachedRegionLocation doesn't have size limit.
> ------------------------------------------------------------
>
> Key: HBASE-723
> URL: https://issues.apache.org/jira/browse/HBASE-723
> Project: Hadoop HBase
> Issue Type: Bug
> Components: client
> Affects Versions: 0.1.3
> Environment: hbase 0.3.0
> Reporter: Jb Lee
> Assignee: Andrew Purtell
> Priority: Minor
>
> cachedRegionLocation stores region locations of tables whenever new region is
> looked up. However, the enties are deleted only when TableServers object is
> closed or locateRegion is called with false useCache argument. Therefore, it
> seems to grow without limit and cause out of memory exception.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
