Connector Lazy Activation leaks managed connections
---------------------------------------------------
Key: GERONIMO-2800
URL: https://issues.apache.org/jira/browse/GERONIMO-2800
Project: Geronimo
Issue Type: Bug
Security Level: public (Regular issues)
Components: connector
Affects Versions: 2.0-M2
Reporter: Kevan Miller
Assigned To: Kevan Miller
Fix For: 2.0-beta1
GERONIMO-2715 added "lazy connection" support to geronimo-connector.
I'm seeing problems where connectors are not being reused properly. The result
is that connectors are not being returned to their pool. Eventually, the pool
of available connections is exhausted and no more connections can be obtained
from the pool.
The basic scenario is:
Connection c = ConnectionFactory.createConnection();
c.close();
c.close();
Connection newC = ConnectionFactory.createConnection();
newC.close();
On the first c.close(), the connectionInfo is disassociated from the
ConnectionProxy. However, on the second c.close(), connectionInfo is
re-associated with the ConnectionProxy. However, since the second close() call
is idempotent, the connection never leaves the pool. When a new Connection is
created, the connection is re-used from the pool, but has two ConnectionInfo's
being tracked as handles. This extra handle prevents the ConnectionInfo from
ever being returned to the pool.
I'm a relative novice when it comes to our Connector implementation. It's
possible that I'm missing another point where the problem could be fixed.
However, I don't see it. I'm also worried about concurrency issues with the
current approach.
I plan on setting lazyConnect to false in the transaction-jta11 and
client-transaction configs. If there aren't ideas on fixing, I think we'll want
to revert the lazy connection code...
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
