[jbpm-users] RE: jBpm 3.2.3 not releasing database connections

[Brian Westrich]

Hello again,

On June 1, I sent a post out reporting a connection release issue with jBPM
3.2.3.  There are no replies to this post.

Today, we realized this same issue is causing databse connections to leak in
the JobExecutorThread (which runs async jobs and timers).

We worked around the initial problem quite easily with a few lines of code
(see below). But resolving it in JobExecutorThread would require a lot more
hackery. Before we hack away, I wanted to again see if anyone has come
across this issue, and if so how you might have addressed it.

Hope to hear from you soon,

Brian

== Original post begins below ==


Hello jbpm gurus,

We're seeing database connection leakage with jBPM 3.2.3 in our unit tests.

Specifically, this JUnit test fails, showing that connections are not
being released back to the connection pool:
       int connectionsBorrowed = datasource.getBorrowedConnectionsCount();
                  // getBorrowedConnectionsCount() is a method on
Oracle's UCP connection pooling
       JbpmConfiguration config =
JbpmConfiguration.parseResource("jbpm.cfg.xml");
       JbpmContext context = config.createJbpmContext();
       context.getSession();
       assertEquals(connectionsBorrowed + 1,
datasource.getBorrowedConnectionsCount()); // passes
       context.close();
       assertEquals("db connection leakage", connectionsBorrowed, datasource
               .getBorrowedConnectionsCount()); // fails

Here's the snippet of our jbpm.cfg.xml file that configures the
persistence service for our test cases:
   <service name="persistence">
       <factory>
               <bean class="org.jbpm
.persistence.db.DbPersistenceServiceFactory">
                       <field name="dataSourceJndiName"><string
value="jdbc/bpsmDS"/> </field>
                               <field name="isCurrentSessionEnabled"><false
/></field>
                               <field name="isTransactionEnabled"><true
/></field>
               </bean>
       </factory>
   </service>

The issue might be related to the DbPersistenceService methods
getConnection(boolean) and getSession(). getConnection(boolean)
initially sets mustConnectionBeClosed to true when we get a connection
out of the datasource, but getSession() then sets this flag back to
false.

As a workaround, we wrote the following custom method for closing a
JbpmContext):
   public void close(JbpmContext context) {
       Connection connection = context.getConnection();
       context.close();
       try {
           connection.close(); // release connection back to pool
       } catch (SQLException e) {
           throw new IllegalStateException(e);
       }
   }

In our non-test environment (where we use JTA transactions), we are
not experiencing connection leakage, so my guess is that the container
(WebSphere) has been closing the connection automatically at the end
of the transaction.

Questions:
 1. Have we misconfigured jBPM or is there some other reason why we're
leaking database connections in our test environment?
 2. Is it ok if we start explicitly calling connection.close() in the
container? For example, will doing so prevent the container from
committing it's transaction (I'm guessing not)? So far our testing
shows no problems whether or not we explicitly call close() in a JTA
environment, but I wanted to make sure about this....


Looking forward to your thoughts on the above questions,

Brian Westrich
bw at mcwest.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.jboss.org/pipermail/jbpm-users/attachments/20090713/e2265d97/attachment.html