My guess would be that the service registry is not being released. Let me look... Brett, that makes me wonder if we ever properly deal with releasing the OSGi classloader stuff... http://lists.jboss.org/pipermail/wildfly-dev/2013-July/000466.html contains the details about deployment level classes are leaking (CDI beanmanager passed into deployment, bean validation factory, others...). In WildFly 8, We do have a META-INF/services/org.hibernate.engine.transaction.jta.platform.spi.JtaPlatformProvider file that contains org.jboss.as.jpa.hibernate4.DefaultJtaPlatform The org.jboss.as.jpa.hibernate4.DefaultJtaPlatform is a system module class but could also be an application deployment class. Question: Do we have a way to unload services that have been loaded when the original classloader (deployment) is no longer needed (e.g. when SessionFactory is closed)? I'm not sure that would help for the leak that I am reproducing which could be a different problem. By the way, this is with Hibernate ORM 4.3.0.Beta3. Scott _______________________________________________ hibernate-dev mailing list hibernate-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/hibernate-dev

Thinking through hibernate-osgi cleanup in general in https://hibernate.atlassian.net/browse/HHH-8378 ----- Original Message ----- From: "Brett Meyer" <brmeyer(a)redhat.com&gt; To: "Steve Ebersole" <steve(a)hibernate.org&gt; Cc: "Scott Marlow" <smarlow(a)redhat.com&gt;, "hibernate-dev" <hibernate-dev(a)lists.jboss.org&gt; Sent: Wednesday, July 24, 2013 11:13:04 AM Subject: Re: [hibernate-dev] report of a WildFly memory leak that may come from hibernate service loader... In hibernate-osgi, one area that should be cleaned up is the static ClassLoaderHelper#overridenClassLoader (used in #withTccl, etc.) -- that's not currently nulled-out when the hibernate-osgi bundle is deactivated. Other than that, I only use BootstrapServiceRegistryBuilder#with to add the OsgiClassLoader. ----- Original Message ----- From: "Steve Ebersole" <steve(a)hibernate.org&gt; To: "Scott Marlow" <smarlow(a)redhat.com&gt; Cc: "hibernate-dev" <hibernate-dev(a)lists.jboss.org&gt; Sent: Monday, July 15, 2013 2:53:40 PM Subject: Re: [hibernate-dev] report of a WildFly memory leak that may come from hibernate service loader... My guess would be that the service registry is not being released. Let me look... Brett, that makes me wonder if we ever properly deal with releasing the OSGi classloader stuff... http://lists.jboss.org/pipermail/wildfly-dev/2013-July/000466.html contains the details about deployment level classes are leaking (CDI beanmanager passed into deployment, bean validation factory, others...). In WildFly 8, We do have a META-INF/services/org.hibernate.engine.transaction.jta.platform.spi.JtaPlatformProvider file that contains org.jboss.as.jpa.hibernate4.DefaultJtaPlatform The org.jboss.as.jpa.hibernate4.DefaultJtaPlatform is a system module class but could also be an application deployment class. Question: Do we have a way to unload services that have been loaded when the original classloader (deployment) is no longer needed (e.g. when SessionFactory is closed)? I'm not sure that would help for the leak that I am reproducing which could be a different problem. By the way, this is with Hibernate ORM 4.3.0.Beta3. Scott _______________________________________________ hibernate-dev mailing list hibernate-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/hibernate-dev _______________________________________________ hibernate-dev mailing list hibernate-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/hibernate-dev