Hi, We are implementing internal notification system for our customer. On some point we have requirement to implement message prioritization. The simplest way, in my opinion, is to use JMS message prioritization feature describer in this doc: http://activemq.apache.org/how-can-i-support-priority-queues.html
In our case sender and reciever is Camel routes. On reviewer side we have 1 conduming endpoing with 2 or more concurrent consumers. In the reciever route we do routing and balancing of messages between external service providers. We can't use many separate endpoints with selective consumers to implement prioritization, because of implementation logic in balancer. We have stand alone Activemq broker for supporting our notification gate. For persistence we try 2 options: 1. Plain JDBC: After some threshold of queue size (about 6000msg) hi priority messages stoped to dispatch. Even after all low priority messages dispatched ! We are see not dispatched messages through web console and JMX. Restarging the broker will dispatch messages. Similar to described in that bug: http://fusesource.com/issues/browse/MB-747 In searching of solution we try to use journaled JDBC... 2. Journaled JDBC: Config: <destinationPolicy> <policyMap> <policyEntries> <policyEntry memoryLimit="1mb" producerFlowControl="true" topic=">"> <pendingSubscriberPolicy> <vmCursor/> </pendingSubscriberPolicy> </policyEntry> <policyEntry queue=">" memoryLimit="10mb" prioritizedMessages="true" producerFlowControl="true" maxAuditDepth="100000"> </policyEntry> </policyEntries> </policyMap> </destinationPolicy> <managementContext> <managementContext createConnector="true"/> </managementContext> <persistenceFactory> <journalPersistenceAdapterFactory dataDirectory="${activemq.base}/data" journalLogFiles="5" dataSource="#mssql-ds"/> </persistenceFactory> After 2200 msgs in queue messages are dispatched without prioritization at all. But the is no stacked messages. After queue size is down to 2200 and less prioritization is restored to expected. Changing log size and other documented params of broker do not resolve issue :( For both variants we are tried all versions from: 5.4.* to latest 5.5.* incude FUSE Source builds. Situation 100% reproducible. Please, help with that... Actualy we already in production with variant 2 and must constantly tune consuming speed to not exceed 2000 msgs in queue. P.S.: currently we are investigating that ticket https://issues.apache.org/jira/browse/AMQ-3436 . Can it be relevant ? -- Regards, Vladimir Konkov
