Michael Samson created AMQ-5188:
-----------------------------------
Summary: AMQ broker throws OutOfMemoryError even when global
flow-control memoryUsage limit is imposed
Key: AMQ-5188
URL: https://issues.apache.org/jira/browse/AMQ-5188
Project: ActiveMQ
Issue Type: Bug
Components: Broker
Affects Versions: 5.8.0
Environment: AMQ 5.8.0 broker running on 64-bit Linux
AMQP Qpid (.26) JMS producers running on 64-bin Linux
Reporter: Michael Samson
Attachments: AMQ_heap1.png, AMQ_heap2.png,
QpidProducerAMQOutOfMemory.java, activemq.xml, qpid-producer.tar
* 5.8.0 broker was configured with persistence=false and a global
<memoryUsage limit="700 mb"/>
* 5.8.0 broker JVM memory was set to default of 1G
* No active consumers
* 10 Qpid (version .26) AMQP producers sent messages to 10 broker queues until
flow control was reached
* Over time the AMQP connections/producers were closed and recreated and they
reattempted to send messages. As this cycle continues, eventually ActiveMQ
threw OutOfMemoryErrors.
I grabbed a heap dump and ActiveMQTextMessages consumed 874M of the heap.
Taking a closer look, it appears that each
org.apache.activemq.broker.region.Queue's messagesWaitingForSpace map grows
unbounded and causes the OutOfMemory.
I will attach a sample maven program that reproduces the problem along with my
activemq configuration, heap screenshots, etc.
Are their any workarounds to this issue? It is a potential use case for our
customers if their queue consumers die for an extended period of time.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
