[
https://issues.apache.org/jira/browse/AMQ-5898?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
James Furness updated AMQ-5898:
-------------------------------
Description:
Hi Tim,
AMQ-5187 breaks some of our message routing where we use virtual queues to
control mirroring/replication of messages.
A simple example...
{code}
VIRTUAL.PUB.ALL
-> queue://SUBSCRIBER1
-> queue://SUBSCRIBER2
VIRTUAL.PUB.SUBSCRIBER1
-> queue://SUBSCRIBER1
{code}
...fails [this
assert|https://github.com/apache/activemq/commit/f55edcfa25de1b55659a7113be60360c531ffa8a#diff-0def63df6ee6c0e4f8adcf00011b2f07R70]
because there are two composite destinations that route to
{{queue://SUBSCRIBER1}}. With asserts disabled messages are routed as expected.
I have attached a test case exemplifying this.
We use layers of composite queues to achieve explicit routing of messages to
either one consumer or all consumers, and (also with static subscriptions) to
target optimal routes across a mixture of LAN and WAN links.
Fully appreciate that subscription recovery from virtual topics of a mapped
queue is a beneficial thing to do, however from our perspective it is also
useful to retain the behaviour of being able to have a many-to-one mapping
between composite queues and physical queues. For our own use case we don't
have any requirement for subscription recovery - we require cast-iron
guarantees around messaging so all messages persistent and are delivered to one
or more physical queues on brokers with persistence enabled, so this obviates
the need for subscription recovery.
Could this validation relaxed, could it be made possible for the new behaviour
to be disabled or do you have any suggestions as to how else we could achieve
our use case?
Thanks,
James
was:
Hi Tim,
AMQ-5187 breaks some of our message routing where we use virtual queues to
control mirroring/replication of messages.
A simple example...
{code}
VIRTUAL.PUB.ALL
-> queue://SUBSCRIBER1
-> queue://SUBSCRIBER2
VIRTUAL.PUB.SUBSCRIBER1
-> queue://SUBSCRIBER1
{code}
...fails [this
assert|https://github.com/apache/activemq/commit/f55edcfa25de1b55659a7113be60360c531ffa8a#diff-0def63df6ee6c0e4f8adcf00011b2f07R70]
because there are two composite destinations that route to
{{queue://SUBSCRIBER1}}. With asserts disabled messages are routed as expected.
We use layers of composite queues to achieve explicit routing of messages to
either one consumer or all consumers, and (also with static subscriptions) to
target optimal routes across a mixture of LAN and WAN links.
Fully appreciate that subscription recovery from virtual topics of a mapped
queue is a beneficial thing to do, however from our perspective it is also
useful to retain the behaviour of being able to have a many-to-one mapping
between composite queues and physical queues.
Could this validation relaxed, could it be made possible for the new behaviour
to be disabled or do you have any suggestions as to how else we could achieve
our use case?
Thanks,
James
> Physical queues forwarded from many virtual destinations no longer supported
> ----------------------------------------------------------------------------
>
> Key: AMQ-5898
> URL: https://issues.apache.org/jira/browse/AMQ-5898
> Project: ActiveMQ
> Issue Type: Bug
> Affects Versions: 5.11.1
> Reporter: James Furness
> Attachments: ActiveManyCompositeQueuesToOnePhysicalQueueTest.java
>
>
> Hi Tim,
> AMQ-5187 breaks some of our message routing where we use virtual queues to
> control mirroring/replication of messages.
> A simple example...
> {code}
> VIRTUAL.PUB.ALL
> -> queue://SUBSCRIBER1
> -> queue://SUBSCRIBER2
> VIRTUAL.PUB.SUBSCRIBER1
> -> queue://SUBSCRIBER1
> {code}
> ...fails [this
> assert|https://github.com/apache/activemq/commit/f55edcfa25de1b55659a7113be60360c531ffa8a#diff-0def63df6ee6c0e4f8adcf00011b2f07R70]
> because there are two composite destinations that route to
> {{queue://SUBSCRIBER1}}. With asserts disabled messages are routed as
> expected.
> I have attached a test case exemplifying this.
> We use layers of composite queues to achieve explicit routing of messages to
> either one consumer or all consumers, and (also with static subscriptions) to
> target optimal routes across a mixture of LAN and WAN links.
> Fully appreciate that subscription recovery from virtual topics of a mapped
> queue is a beneficial thing to do, however from our perspective it is also
> useful to retain the behaviour of being able to have a many-to-one mapping
> between composite queues and physical queues. For our own use case we don't
> have any requirement for subscription recovery - we require cast-iron
> guarantees around messaging so all messages persistent and are delivered to
> one or more physical queues on brokers with persistence enabled, so this
> obviates the need for subscription recovery.
> Could this validation relaxed, could it be made possible for the new
> behaviour to be disabled or do you have any suggestions as to how else we
> could achieve our use case?
> Thanks,
> James
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
