hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Lars George (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-7035) thrift server closes HTable of open Scanners
Date Sat, 13 Jul 2013 20:15:48 GMT

    [ https://issues.apache.org/jira/browse/HBASE-7035?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13707828#comment-13707828
] 

Lars George commented on HBASE-7035:
------------------------------------

Connections are reference counted, see [HConnectionManager|https://github.com/apache/hbase/blob/0.94/src/main/java/org/apache/hadoop/hbase/client/HConnectionManager.java#L296]
for reference. So given the issues we had with connection identities in there are fixed, we
should be OK to close a table as the connection will be counted down? But then the scanner
is a non-counted reference, but only when all tables are closed and the reference count ends
up being zero it gets really closed. By that time scanners should be all done?

The other option we discussed is to hold on to the table for the scanner, but then we have
one for every instance, which is way too much under load. We only need to hold on to one per
table, therefore keeping the umbilical cord alive (so to speak).
                
> thrift server closes HTable of open Scanners
> --------------------------------------------
>
>                 Key: HBASE-7035
>                 URL: https://issues.apache.org/jira/browse/HBASE-7035
>             Project: HBase
>          Issue Type: Sub-task
>          Components: Thrift
>    Affects Versions: 0.94.4
>            Reporter: Devin Bayer
>              Labels: thrift2
>         Attachments: old-hbase-thrift-v1.patch
>
>   Original Estimate: 4h
>  Remaining Estimate: 4h
>
> ThriftHBaseServiceHandler.openScanner() does this:
> 1. table = pool.getTable()
> 2. scanner = table.getScanner()
> 3. table.close()
> 4. return scanner
> While back porting the thrift server to 0.92.6, I found that table.close() calls connection.close().
Further calls to scanner.next() raise a ConnectionClosed exception. The unit tests do not
catch this since they reuse an open HConnection instance.
> This might work on trunk, but depends on the implementations of HTablePool, HTable and
HConnectionManager. Even with the pool wrapper, if the pool is full, table.close() may be
called, which may invalidate the table. Also,  HTable is not thread-safe, but they are being
reused since they go back in the pool.
> I suggest storing the table handle along with the scanner.

--
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

Mime
View raw message