Cxf Failover causes leak memory on periodic http requests

[LORENZA Adnan <ad...@gmail.com>]

Hi,

I developed  a camel cxf component that is configured to failover to
alternate addresses in case of connections /availability failures.
The camel component is a simple timer that sends every second a request to
a webservice.

The camel route is the following :

public void configure() throws Exception {


from("timer:timerRetrieveAvailableJobs?fixedRate=true&period=1000")
            .setBody(constant(getRequestMessage())
            .setHeader(CxfConstants.OPERATION_NAME,
constant("getAvailableJobs"))

.to("cxf:bean:cxfJobsWsEndpoint?synchronous=true&loggingFeatureEnabled=true");

}

private String getRequestMessage(){
            //Build the right request
             ...
}

The camel-context.xml configures the clustering failover as below :
...
<cxf:cxfEndpoint id="cxfJobsWsEndpoint"
serviceClass="net.jobs.ws.myPTServiceClass" address="
http://localhost/myservice/JobWS">
   <cxf:features>
            <clustering:failover>
               <clustering:strategy>
                        <bean
class="org.apache.cxf.clustering.SequentialStrategy">
                           <property name="alternateAddresses">
                             <util:list>
                                    <value>http://server1/JobWS</value>
                                    <value>http://server2/JobWS</value>
                             </util:list>
                          </property>
                        </bean>
               </clustering:strategy>
            </clustering:failover>
  </cxf:features>
 <cxf:properties>
            <entry key="dataFormat" value="MESSAGE" />
            </cxf:properties>
</cxf:cxfEndpoint>
...

After a few hours, a dump of JVM memory shows that some objects of CXF
(org.apage.cxf.xxxx) are created and remain in the memory. They are not
accessible by the GC. Their number continues to grow and causes a memory
leak after a long term :

command :  jmap -histo:live 25698 | grep cxf
  8:          90660         9428640
[Lorg.apache.cxf.phase.PhaseInterceptorChain$InterceptorHolder;
  20:        135989      3263736
org.apache.cxf.phase.PhaseInterceptorChain$InterceptorHolder
  22:         45328        2900992  org.apache.cxf.message.MessageImpl
  24:         45330        2175840
org.apache.cxf.phase.PhaseInterceptorChain
  25:         22664        1994432  org.apache.cxf.message.ExchangeImpl
  33:         45328        1087872
org.apache.cxf.phase.PhaseInterceptorChain$PhaseInterceptorIterator
  36:         22664         906560
org.apache.cxf.clustering.FailoverTargetSelector$InvocationContext
  40:         22664         725248
org.apache.cxf.helpers.LoadingByteArrayOutputStream$1
  41:         45328         725248  org.apache.cxf.binding.soap.SoapMessage
  46:         22664         543936
org.apache.cxf.message.MessageContentsList
  51:         22664         362624
org.apache.cxf.helpers.LoadingByteArrayOutputStream
  52:         22664         362624
org.apache.cxf.clustering.FailoverTargetSelector$InvocationKey

I tried the same test without failover configuration and I was surprised :
 the component continues to work without memory problem. The objects above
are collected and removed by the GC.
It seems that the classes that implement the clustering failover retain
created objects and prevent their removal.

The test has been done on the following configurations :

  *  apache-servicemix-3.4.0,  cxf 2.4.4, camel 2.8.3

  *  apache-servicemix-3.5.0-fuse-00-00, cxf 2.2.11, camel 2.5

Any help will be appreciated.
Thanks

Regards
 Adnan