[ https://issues.apache.org/jira/browse/AMQ-6976?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16493417#comment-16493417 ]
Markus L. commented on AMQ-6976: -------------------------------- We are using the RAR Connection Pool approach and therefore the connections will stay open "forever" and are reused by several processes. May this be a problem? > GC goes crazy under load. Heap is filled up with TransportConnectionState > objects > --------------------------------------------------------------------------------- > > Key: AMQ-6976 > URL: https://issues.apache.org/jira/browse/AMQ-6976 > Project: ActiveMQ > Issue Type: Bug > Components: Broker > Affects Versions: 5.15.3 > Environment: Windows Server 2012 R2 - 64bit > 230 GB RAM > SSD HD > Reporter: Markus L. > Priority: Critical > Attachments: image-2018-05-28-13-54-21-048.png, > image-2018-05-28-16-23-36-564.png > > > We are running a so called "Shared Nothing" setup with several Wildfly 8.2.1 > instances (10) on several host machines (4)- all connected to one standalone > Activemq installation - Normally we have 600-700 connections in total to our > broker. > The Activemq installation is configured to use 8GB RAM. > {code:java} > <broker xmlns="http://activemq.apache.org/schema/core" brokerName="dcdng" > id="dcdng" useJmx="true" useShutdownHook="false" > schedulerSupport="false" dataDirectory="C:/activemq/data" > advisorySupport="false"> > <managementContext> > <managementContext createConnector="true" connectorPort="61610" > suppressMBean="endpoint=dynamicProducer,endpoint=Consumer,connectionName=*,destinationName=ActiveMQ.Advisory.*"/> > </managementContext> > <!-- ... --> > </broker>{code} > The memory usage is set to 500mb which is sufficient since our messages are > relative small. (MaxMessageSize: 41442) > > The system performs great until we put it under high load. Then the heap > usage grows and it seems the memory cannot get cleaned up properly. > !image-2018-05-28-16-23-36-564.png! > At the end I took a heap dump and the MAT prints the following: > !image-2018-05-28-13-54-21-048.png! > So why are there so many TransportConnectionState objects left an not cleaned > up properly. When the system performs under "normal" load, everything seems > to be fine. -- This message was sent by Atlassian JIRA (v7.6.3#76005)