activemq-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Gary Tully (JIRA)" <>
Subject [jira] Commented: (AMQ-1730) Bad use of Jms field JMSXDeliveryCount. Related to RedeliveryCount and message prefetch
Date Fri, 07 Nov 2008 16:32:12 GMT


Gary Tully commented on AMQ-1730:

On the meaning of JMSXDeliveryCount, note that the semantics of redelivery indicators in JMS
are not absolute. They are in indication only that redelivery may have occurred or that a
transaction rolledback. 
The fact that redispatch of prefetched messages on subscription close, increments the redelivery
count, is reasonable w.r.t the spec. The extent of this can be managed by the prefetch value
for the subscription.

> Bad use of Jms field JMSXDeliveryCount. Related to RedeliveryCount and message prefetch

> ----------------------------------------------------------------------------------------
>                 Key: AMQ-1730
>                 URL:
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Broker
>    Affects Versions: 5.0.0, 5.1.0
>            Reporter: Alexis Kinsella
>            Assignee: Rob Davies
>             Fix For: 5.3.0
> JMSXDeliveryCount  has to be incremented only on transactional delivery failure ( RuntimeException
on processing message, ... ).
> JMSXDeliveryCount  is actualy used in correlation with field: 'redeliveryCounter'.
> But RedeliveryCounter  field is used to be incremented each time the message is preteched
or sent to a consumer => It does not mean that message has been processed by business in
transaction with a failure. It just has been prefetched by consumer ou subscriber. 
> A 'not consumed message' can be give back to the broker when the consumer is closed by
user, because it has been prefetched but not really consumed!
> It does not match the meaning of the JMS field JMSXDeliveryCount  :
> "If a failure occurs within transactional processing then the JMSXDeliveryCount is incremented".
> JMSXDeliveryCount  field only has to be incremented on rollback (isn't it? ). This is
why Message.redeliveryCount can not be used. Or the behavour of field Message.redeliveryCount
has to be changed.
> You can either :
> * create a new counter on message incremented only on rollback, or
> * modify classes : and to remove redeliveryCount
increment and increment it only on rollback.
> *
>                     if (inAckRange) {
> //                        node.incrementRedeliveryCounter();
>                         if (ack.getLastMessageId().equals(messageId)) {
>                             destination = node.getRegionDestination();
>                             callDispatchMatched = true;
>                             break;
>                         }
>                     }
> *
>                 for (MessageReference ref : sub.remove(context, this)) {
>                     QueueMessageReference qmr = (QueueMessageReference)ref;
> //                    qmr.incrementRedeliveryCounter();
>                     if( qmr.getLockOwner()==sub ) {
>                         qmr.unlock();
> //                        if (!qmr.isDropped() && !qmr.isAcked()) {
> //                        	qmr.incrementRedeliveryCounter();
> //                        }
>                     }
>                     list.add(qmr);
>                 }
> BTW, In this code it seems there is a second bug, in case of  test "qmr.getLockOwner()==sub"
is true qmr is incremented a second time ?! Is it right ?
> The result of this problem is the following: 
> With Spring and DefaultMessageListenerContainer, a message is consumed one by one. This
is why a message prefteched many times, on first real consuming has a JMSXDeliveryCount  with
high value not reflecting the reality.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message