[
https://issues.apache.org/jira/browse/AMQ-4533?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13690424#comment-13690424
]
Gary Tully commented on AMQ-4533:
---------------------------------
However, if there are very few messages and lots of consumers, using a prefetch
> 0 will not have much impact. Prefetch is really only beneficial when
consumers are very fast and when there are lots of messages in the queue or
many producers. When consumers take a long or variable amount of time, it is
best to leave the messages in the queue (prefecth=1) for other consumers to
access.
ok, we may need to make the slowness determination some sort of strategy.
Have a thought on how best to determine the 'slowness" of a consumer, maye time
since last ack.
It needs to a fast calculation. The decision to dispatch based on prefetch was
already firing an advisory for a slow consumer so that was reused for the
current abort policy.
I would be interested to know if you can get the test to fail with a
prefetch=1, also note that just the consumer is aborted, not the connection,
abortConnection=false (Default)
> Messages stuck in queue with redelivered=true
> ---------------------------------------------
>
> Key: AMQ-4533
> URL: https://issues.apache.org/jira/browse/AMQ-4533
> Project: ActiveMQ
> Issue Type: Bug
> Components: JMS client
> Affects Versions: 5.7.0
> Environment: Fuse Message Broker 5.7.0
> Reporter: Jason Shepherd
> Attachments: AMQ4533Test.java, AMQ4533TestPatch.txt,
> AMQ4533TestPatch.txt, AMQ4533TestPatch.txt, kahaPendingMessages.zip
>
>
> We're getting message stuck in queues with the
> redelivery flag set to true.
> We used the following test model: put every 1 second 50 messages
> sequentially, and after that, the rest of 1000 msgs quickly to INPUT_QUEUE
> and
> while starting 25 listeners cosuming from INPUT_QUEUE, which takes about 30
> seconds to move the message to RECEIPT_QUEUE, 10 other listeners on
> RECEIPT_QUEUE consume and counts them.
> We tried making one of the consumer slow by setting the
> processing time to 100000 seconds (sleep) and putting a heavy load in
> 500 threads every 1 ms to some other queues the same time.
> Our test case is attached, you might need to install some dependencies
> to the local maven repository manually:
> mvn install:install-file -DgroupId=org.apache.activemq
> -DartifactId=activemq-core -Dversion=5.7.0-fuse-71-047 -Dpackaging=jar
> -Dfile=activemq-core-5.7.0.fuse-71-047.jar
> mvn install:install-file -DgroupId=org.apache.kahadb
> -DartifactId=kahadb -Dversion=5.7.0-fuse-71-047 -Dpackaging=jar
> -Dfile=kahadb-5.7.0.fuse-71-047.jar
> mvn install:install-file
> -DgroupId=org.apache.geronimo.management.specs
> -DartifactId=geronimo-j2ee-management_1.1_spec -Dversion=1.0.1
> -Dpackaging=jar -Dfile=geronimo-j2ee-management_1.1_spec-1.0.1.jar
> mvn install:install-file -DgroupId=org.apache.activemq.pool
> -DartifactId=activemq-pool -Dversion=5.7.0-fuse-71-047 -Dpackaging=jar
> -Dfile=activemq-pool-5.7.0.fuse-71-047.jar
> To run the test, simply use the Maven test target:
> mvn clean test
> If the problem occurs the you'll get a message like this in the test
> results, (target/surefire-reports):
> java.lang.AssertionError: Still messages in InputQueue expected:<0>
> but was:<365>
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
