Fabian González created AMQ-6775:
------------------------------------
Summary: Race condition makes messages queued in the session be
delivered before messages queued in
Key: AMQ-6775
URL: https://issues.apache.org/jira/browse/AMQ-6775
Project: ActiveMQ
Issue Type: Bug
Components: JMS client
Affects Versions: 5.14.3, 5.16.0
Reporter: Fabian González
I found this situacion in org.apache.activemq.ActiveMQSessionExecutor.iterate:
As I understand, the idea is that if there are messages queued on the consumers
they are delivered to the listeners and if that is not the case, you should
dispatch the messages in the session:
{code:java}
public boolean iterate() {
// Deliver any messages queued on the consumer to their listeners.
for (ActiveMQMessageConsumer consumer : this.session.consumers) {
if (consumer.iterate()) {
return true;
}
}
// No messages left queued on the listeners.. so now dispatch messages
// queued on the session
MessageDispatch message = messageQueue.dequeueNoWait();
if (message == null) {
return false;
} else {
dispatch(message);
return !messageQueue.isEmpty();
}
}
{code}
Now I the following race condition arises:
1) thread A (ActiveMQ Session Task) invokes .ActiveMQSessionExecutor
2) When this part of the code is executed:
{code:java}
for (ActiveMQMessageConsumer consumer : this.session.consumers) {
if (consumer.iterate()) {
return true;
}
}
{code}
ActiveMQMessageConsumer.iterate is invoked. There are messages unconsumed in
the consumer, but as unconsumedMessages is not started, a null is returned as
if there were no messages queued.
3) Thread A is interrupted
4) Thread B (ActiveMQConnection[xx]Scheduler) invokes
ActiveMQMessageConsumer.start, unconsumed messages are started.
5) Thread A continues to deliver the messages queued in the session (*notice
that if thread B starts unconsumed messages before thread A which happens most
of times the messages in the consumer queue would have been dispatched*
6) Thread A dispatch a message from the session when there are messages from
the consumer pending.
This race condition makes makes that in some cases a message which has been
rollbacked and get queued in the consumer queued is processed *after* another
message in the session consumer (which was enqueued after the former message).
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
