Matthew Stratton created AMQ-7078:
-------------------------------------
Summary: possibly unintended competing consumer behavior with
virtual topic consumer queues using wildcards
Key: AMQ-7078
URL: https://issues.apache.org/jira/browse/AMQ-7078
Project: ActiveMQ
Issue Type: Bug
Affects Versions: 5.15.6
Reporter: Matthew Stratton
Attachments: VirtualTopicWildcardMisrouting.java,
gtully_irc_conversation.txt
I spoke with [~gtully] over irc about this, attached a log of that and a unit
test demonstrating what I mean attached. Verified in 5.15.6, likely an issue as
far back as you go, but hard to verify due to a related bug AMQ-6443.
If you use virtual topic consumer queues using wild wildcards and non-unique
consumer queue prefixes (the part before "VirtualTopic" in the queue name) you
can end up with duplicate messages in the wildcard queue while taking messages
from nonwildcard queues. E.g. "Consumer.foo.VirtualTopic.>" and
"Consumer.foo.VirtualTopic.bar".
"Consumer.foo.VirtualTopic.>" gets dispatched messages sent to the topic
"VirtualTopic.bar" but consumers of "Consumer.foo.VirtualTopic.>" also become
consumers of *any virtual topic consumer queue that matches the wildcard
pattern*. So consumers of wildcard virtual topic queues can receive messages
multiple times while depriving others of the messages they expect. If 5
messages are sent to "VirtualTopic.bar", 5 messages are dispatched to both
"Consumer.foo.VirtualTopic.>" and "Consumer.foo.VirtualTopic.bar", this is
correct so far. But the "Consumer.foo.VirtualTopic.>" consumer will end up with
more than 5 messages (being detected as duplicated if the client supports it)
while the "Consumer.foo.VirtualTopic.bar" ends up with less than it expects.
Possibly related to https://issues.apache.org/jira/browse/AMQ-6643, though i
think the fix (and me trying to verify its working as expected) simply made the
behavior obvious. I dont think the AMQ-6643 fix introduced this behavior. I
dont think this is a bug, but an unintended behavior of wildcards between
virtualtopic destination mapping and queue wildcards in general. The later
would need to explicitly guard against the case.
There is an obvious and simple client side work around (dont use the same
prefix) but at the very least I think this should be called out in
documentation as it surprised me and I am quite familiar with how AMQ and
virtual topics work. It makes sense in hindsight, but the behavior is very
subtle if you arent paying attention and requires you to connect some dots.
Most of my amq clients are outside of my direct control and managed by people
with little to no amq/jms experience so the simple client side work around is
less simple.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
