Dead lock during XA recovery when activeMq 5.0.0 is used with Atomikos
----------------------------------------------------------------------
Key: AMQ-1569
URL: https://issues.apache.org/activemq/browse/AMQ-1569
Project: ActiveMQ
Issue Type: Bug
Affects Versions: 5.0.0
Reporter: Mathieu Baril
Hi,
I developed an application that runs in an OSGI environment (Equinox). My
application is developed using Spring Dynamic Modules 1.0-rc2, ActiveMq 5.0.0
and Atomikos 3.2.2.
Sometimes (50% of the time) when I start my application, the bundle that
declare the connection factories to the jms broker runs into deadlock.
This is the declaration of my connection factories:
<bean id="activemq.connectionFactory"
class="com.trilliantnetworks.griffin.utils.CustomXAJMSConnectionFactory">
<property name="brokerURL" value="
failover:(tcp://software2:61616,tcp://software1:61616)" />
</bean>
<!-- Configure the JMS connector; call init to register for recovery! -->
<bean id="atomikos.connectionFactory"
class="com.atomikos.jms.TopicConnectionFactoryBean" init-method="init">
<property name="resourceName" value=" jms.connectionFactory"/>
<property name="xaTopicConnectionFactory" ref=" activemq.connectionFactory
"/>
</bean>
<bean id="listen.jms.spring.singleConnectionFactory"
class="org.springframework.jms.connection.SingleConnectionFactory"
destroy-method="destroy">
<constructor-arg ref="atomikos.connectionFactory "/>
</bean>
Looking deeper in the code I found that the application block when init is
called in com.atomikos.jms.TopicConnectionFactoryBean.
The init method calls the method isSameRM of
org.apache.activemq.TransactionContext that eventually calls waitForBrokerInfo
in org.apache.activemq.ActiveMQConnection. The method blocks on
brokerInfoReceived.await(). So I tried to found why
brokerInfoReceived.countDown() is never called...
I found that this is because createTransportListener of
org.apache.activemq.transport.failover.FailoverTransport is sometimes called
before the member variable transportListener was set. createTransportListener
is called by the Thread ActiveMQ Transport: tcp://software2/192.9.200.182:61616
while setTransportListener is called by the main Thread that initialize my
bundle.
I am unable to reproduce the issue with activeMQ 4.1.1 so I think that is issue
has been introduced with version 5.0.0.
Also, when I use the connection URL "tcp://software2:61616" instead of
"failover:(tcp://software2:61616,tcp://software1:61616)" my application always
start correctly.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
