[
https://issues.apache.org/jira/browse/ARTEMIS-1937?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16629029#comment-16629029
]
Justin Bertram commented on ARTEMIS-1937:
-----------------------------------------
The nuance here is that each protocol implementation on the broker (e.g. STOMP,
AMQP, MQTT) is responsible for creating addresses & queues which don't exist,
but the core protocol itself never auto-creates addresses or queues. When the
message is routed to the divert the divert has no idea about the originating
protocol, and it doesn't perform any auto-creation on its own. It's possible
the semantics here could be changed but it will take some careful thought and
potentially some significant refactoring.
> Diverts do not work with auto created queues
> --------------------------------------------
>
> Key: ARTEMIS-1937
> URL: https://issues.apache.org/jira/browse/ARTEMIS-1937
> Project: ActiveMQ Artemis
> Issue Type: Bug
> Reporter: Lionel Cons
> Priority: Major
>
> Consider the following configuration snippet:
> {code}
> <address-settings>
> <address-setting match="/queue/test.#">
> <default-address-routing-type>ANYCAST</default-address-routing-type>
> <default-queue-routing-type>ANYCAST</default-queue-routing-type>
> </address-setting>
> <address-setting match="/topic/test.#">
> <default-address-routing-type>MULTICAST</default-address-routing-type>
> <default-queue-routing-type>MULTICAST</default-queue-routing-type>
> </address-setting>
> ...
> </address-settings>
>
> <diverts>
> <divert name="vq-foo-aaa">
> <address>/topic/test.foo</address>
> <forwarding-address>/queue/test.vq.aaa.foo</forwarding-address>
> <exclusive>false</exclusive>
> </divert>
> <divert name="vq-foo-bbb">
> <address>/topic/test.foo</address>
> <forwarding-address>/queue/test.vq.bbb.foo</forwarding-address>
> <exclusive>false</exclusive>
> </divert>
> ...
> </diverts>
> {code}
> The goal of the {{<address-settings>}} part is to make sure that STOMP
> destinations like {{/queue/test.\*}} act like a queue and the ones like
> {{/topic/test.\*}} act like a topic. It works fine and sending a message to a
> non existing destination works as expected.
> The goal of the {{<diverts>}} part is to emulate ActiveMQ 5 virtual
> destinations. A message sent to {{/topic/test.foo}} will get duplicated and
> will end up both in {{/queue/test.vq.aaa.foo}} and in
> {{/queue/test.vq.bbb.foo}}.
> The problem is that the configuration above does not work as a message being
> sent to {{/topic/test.foo}} is lost if the final queue does not exist. OTOH,
> a message directly sent to {{/queue/test.vq.aaa.foo}} will auto create the
> queue and will be kept.
> Adding the following:
> {code}
> <addresses>
> <address name="/queue/test.vq.aaa.foo">
> <anycast>
> <queue name="/queue/test.vq.aaa.foo"/>
> </anycast>
> </address>
> <address name="/queue/test.vq.bbb.foo">
> <anycast>
> <queue name="/queue/test.vq.bbb.foo"/>
> </anycast>
> </address>
> ...
> <addresses>
> {code}
> makes Artemis work as expected but this voids the use of the
> {{<address-settings>}} part.
> IMHO, a message sent to a divert pointing to an address (whether it already
> exists or not) should have the exact same behavior as a message sent directly
> to the forwarding address.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
