hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Eugene Kirpichov (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HBASE-1604) HBaseClient.getConnection() may return a broken connection without throwing an exception
Date Wed, 15 Jul 2009 12:39:14 GMT

    [ https://issues.apache.org/jira/browse/HBASE-1604?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12731415#action_12731415

Eugene Kirpichov commented on HBASE-1604:

2 weeks passed without reaction. Hey, developers!

I've implemented the suggested fix and it improved things somewhat; the described bug is fixed,
and now the situation where a broken connection stays in pool occurs extremely rarely, but
it still occurs. I still don't know why.

> HBaseClient.getConnection() may return a broken connection without throwing an exception
> ----------------------------------------------------------------------------------------
>                 Key: HBASE-1604
>                 URL: https://issues.apache.org/jira/browse/HBASE-1604
>             Project: Hadoop HBase
>          Issue Type: Bug
>          Components: client
>    Affects Versions: 0.19.2
>            Reporter: Eugene Kirpichov
> Consider the code of HBaseClient.getConnection():
> {code}
>     connection.setupIOstreams();
>     return connection;
>   }
> {code}
> Now consider the setupIOstreams() method:
> {code}
>       } catch (IOException e) {
>         markClosed(e);
>         close(); // Removes the connection from pool
>       }
> {code}
> So, if something goes wrong inside of setupIOstreams, then after its invocation the connection
will be broken (will have its .in and .out streams nulls, for example) and will not be in
pool, but will still be returned from the getConnection method and cause further harm (for
example, cause a NullPointerException in further calls such as sendCall, which use the in
and out streams).
> Suggested fix: make the setupIOstreams method rethrow the IOException inside that catch
> Reproduction: Restart the hbase master and/or regionserver while a client program is
running, and put a breakpoint into that catch block.
> I actually observed a situation where the broken connection stayed in the pool, but I
don't yet know how to reproduce it or what is the reason. I am investigating the issue, but
for now at least the aforementioned bug should be fixed.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message