[
https://hibernate.onjira.com/browse/HHH-7020?page=com.atlassian.jira.plug...
]
Shawn Clowater commented on HHH-7020:
-------------------------------------
The first session is opened without a connection and ends up with the logical connection @
1640 and the ResourceRegistry @ 1644
When the connection() method is called for the second session it creates another
LogicalConnectionImpl when it builds the JdbcCoordinatorImpl when creating the
TransactionCoordinatorImpl. However, we now have nested the LogicalConnectionImpl of the
first session as the PhysicalConnection of the second session's LogicalConnectionImpl
@ 1702 with it's own ResourceRegistry @ 1708
When the PreparedStatment registers itself when it is prepared it actually calls down
through both LogicalConnectionImpls and registers with both ResourceRegistries, when it is
released it only gets removed from the second session's ResourceRegistry but is left
in an invalid state in the first session's registry.
When the first session tries to close then it hits the invalid registered resource and
then blows itself out. I'll attach a unit test that shows that the first session has
registered resources but doesn't actually execute any statements of its own.
Connection leak with nested sessions
------------------------------------
Key: HHH-7020
URL:
https://hibernate.onjira.com/browse/HHH-7020
Project: Hibernate ORM
Issue Type: Bug
Components: core, entity-manager
Affects Versions: 4.0.0.Final, 4.0.1, 4.1.0
Environment: Windows 7.
Java 1.6.0_25, Java 1.7.0
Reporter: Zoltán Holub
Attachments: HHH-7020Debug.png, HHH-7020.diff, testcase-connectionleak.zip
I'am using a Hibernate interceptor to track data modifications. In the
interceptor's onFlushDirty() method, i open a new session without interceptors using
the same JDBC connection.
After closing this nested session and the original entitymanager, the JDBC connection
doesn't released.
I could reproduce this problem outside any interceptor. I have made a junit test without
interceptors which shows the same problem.
I have found that Hibernate throws and catches a HibernateException with message
"proxy handle is no longer valid". This exception is logged on debug level.
I investigated that the problem is in the following code:
LogicalConnectionImpl.java
public Connection close() {
LOG.trace( "Closing logical connection" );
Connection c = isUserSuppliedConnection ? physicalConnection : null;
try {
releaseProxies();
jdbcResourceRegistry.close();
if ( !isUserSuppliedConnection && physicalConnection != null ) {
releaseConnection();
}
return c;
}
finally {
// no matter what
physicalConnection = null;
isClosed = true;
LOG.trace( "Logical connection closed" );
for ( ConnectionObserver observer : observers ) {
observer.logicalConnectionClosed();
}
observers.clear();
}
}
The invocation of jdbcResourceRegistry.close() throws this exception, and this is why it
skips the following releaseConnection(). The condition of IF statement is true.
This code is new to Hibernate 4. I have tried 4.0.0, 4.0.1 and 4.1.0-SNAPSHOT and all of
them has the problem. Hibernate 3.6.0 is not affected.
My JUnit test could run with Hibernate 3.x with minor changes. (3.x code also in the
JUnit test)
I have tried different ways to work around the problem. The only way it worked is to use
a simple doWork() and native JDBC operations. This test case also in the unit test.
--
This message is automatically generated by JIRA.
For more information on JIRA, see:
http://www.atlassian.com/software/jira