[ https://issues.apache.org/jira/browse/CASSANDRA-6541?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13861619#comment-13861619 ]
Jonathan Ellis commented on CASSANDRA-6541: ------------------------------------------- A bit more context: # We have heap dumps full of javax.management.remote.rmi.RMIConnectionImpl$CombinedClassLoader # CombinedClassLoader appears to be sandboxing classes used by RMI/JMX and thus instantiates its own copies for each connection # If you only use a stable JMX connection pool you are fine, otherwise you "leak" CCL roots # If you have bad enough old-gen fragmentation to force periodic STW GC, you are also fine; perversely, this will only bite you if CMS is working as designed > Add JVM_OPTS="$JVM_OPTS -XX:+CMSClassUnloadingEnabled" to cassandra-env.sh > -------------------------------------------------------------------------- > > Key: CASSANDRA-6541 > URL: https://issues.apache.org/jira/browse/CASSANDRA-6541 > Project: Cassandra > Issue Type: Bug > Components: Config > Reporter: jonathan lacefield > Priority: Minor > > Newer versions of Oracle's Hotspot JVM , post 6u45, are experiencing issues > with GC and JMX where heap slowly fills up overtime until OOM or a full GC > event occurs, specifically when CMS is leveraged. Running repair exacerbates > this issue. The configuration added to the Summary line helps alleviate this > behavior and should be included in the C* config files by default. -- This message was sent by Atlassian JIRA (v6.1.5#6160)