com.isomorphic.log.RevolvingMemoryAppender memory leak

[dbscott525]

I am using 3.1-p20130910

I have an application for which I was experiencing a memory leak that caused me run out of memory after running it for about an hour. I finally purchased a product called Plumbr (www.plumbr.eu). After 15 minutes of analysis if found a memory leak in the Isomorphic server code. Here is the report I received from Plumbr:

Currently 242 leaked objects exist of class
com.isomorphic.log.LogEntry

The objects are created at 
com.isomorphic.log.RevolvingMemoryAppender.append(org.apache.log4j.spi.LoggingEvent):81

and are being held
in elementData of java.util.ArrayList
in list of java.util.Collections$SynchronizedRandomAccessList
in value of org.apache.commons.collections.map.AbstractLinkedMap$LinkEntry
in data of org.apache.commons.collections.map.LinkedMap
in logs of com.isomorphic.log.RevolvingMemoryAppender
in com.isomorphic.log.RevolvingMemoryAppender
in elementData of java.util.Vector
in appenderList of org.apache.log4j.helpers.AppenderAttachableImpl
in aai of org.apache.log4j.Logger
in parent of org.apache.log4j.Logger
in value of java.util.Hashtable$Entry
in table of java.util.Hashtable
in ht of org.apache.log4j.Hierarchy
in repository of org.apache.log4j.spi.DefaultRepositorySelector
in repositorySelector of org.apache.log4j.LogManager
in elementData of java.util.Vector
in classes of sun.misc.Launcher$AppClassLoader
in other kind of root reference (please let us know if you see this)

[Isomorphic]

An automated tool that knows that an object has been leaked is pretty amazing. How does it know that the system will never access that object again? :)

The fact that RevolvingMemoryAppender has a bunch of LogEntry objects is not a leak - the whole point of this class is to hold up to "maxEntries" LogEntry objects. The value of maxEntries can be configured in log4j.isc.config.xml and we ship with a default of 500.

This report also doesn't pass the "sniff test" that applications using this subsystem have been left running for months under load without running out of memory.