[
https://issues.apache.org/jira/browse/CXF-2756?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12855366#action_12855366
]
Gregor B. Rosenauer commented on CXF-2756:
------------------------------------------
We tested the same service but with a single endpoint that publishes one large
WSDL containing all services.
This setup worked fine, no memory leak was observed (const. around 125MB with
1GB Heap) and performance was stable.
So it's definitely a CXF-issue with a setup that consists of many (>20) Service
Endpoints that each expose their own WSDL...
> OutOfMemoryError with many service endpoints
> --------------------------------------------
>
> Key: CXF-2756
> URL: https://issues.apache.org/jira/browse/CXF-2756
> Project: CXF
> Issue Type: Bug
> Components: Bus
> Affects Versions: 2.2.5
> Environment: Windows 7 x64 Sun JDK 1.6.0_17-b04), Java HotSpot(TM)
> 64-Bit Server VM (build 14.3-b01, mixed mod;
> IBM J9 JDK 64bit IBM J9 VM (build 2.4, J2RE 1.6.0 IBM J9 2.4 AIX ppc64-64
> jvmap6460sr5
> JBoss 5.1.0.GA + JBossWS-CXF 3.2.2.GA
> Reporter: Gregor B. Rosenauer
> Attachments: memory-usage_cxf.png, memory-usage_metro.png,
> stresstest-ear.ear
>
>
> I am deploying an EAR in JBoss which exposes many (approx. 120) endpoints,
> each of those publish a simple WSDL (2 services with 1 parameter) - demo
> attached, if necessary I can add a private attachment with the real world EAR.
> When invoking these service endpoints, CXF always scans the entire EAR and
> (re)builds the service endpoint registry.
> Performance with many (10-20) concurrent service calls degrades and memory
> consumption increases steadily, ultimately leading to an OutOfMemoryException
> in JBoss. It seems the endpoints cannot be found in the cache and so the
> entire registry is rebuilt.
> In a simple test it showed that simply displaying the service endpoints
> exposed WSDL is sufficient to trigger the problematic behaviour:
> {code:title=JBoss server.log|borderStyle=solid}
> 2010-04-06 10:47:34,861 DEBUG
> [org.springframework.beans.factory.support.DefaultListableBeanFactory]
> Invoking init method 'publish' on bean with name 'Vvpvzfpr'
> 2010-04-06 10:47:34,862 INFO
> [org.apache.cxf.service.factory.ReflectionServiceFactoryBean] Creating
> Service {http://vvpvzfpr.vvp.ta2mig.itsv.at/}VvpvzfprService from class
> at.itsv.ta2mig.vvp.vvpvzfpr.IVvpvzfpr
> 2010-04-06 10:47:34,868 INFO [org.apache.cxf.endpoint.ServerImpl] Setting
> the server's publish address to be
> http://127.0.0.1:8080/ta2mig-vvptest1-ear-ta2mig-vvptest1-service-BUILD/Vvpvzfpr
> 2010-04-06 10:47:34,871 INFO
> [org.apache.cxf.common.injection.ResourceInjector] failed to resolve resource
> at.itsv.ta2mig.vvp.vvpvzfpr.Vvpvzfpr/connectionFactory
> 2010-04-06 10:47:34,872 DEBUG
> [org.springframework.beans.factory.support.DefaultListableBeanFactory]
> Finished creating instance of bean 'Vvpvzfpr'
> {code}
> This is for every service at every call, please ignore the "failed to resolve
> resource ..." as the datasource is not available in the local test
> environment.
> Metro does not show this behaviour and reacts very fast with minimum memory
> consumption, but we cannot easily switch the entire web service stack at this
> time and I don't believe CXF has such a problem with many service
> endpoints... however since even calling the WSDL from JBoss' web interface
> triggers the problem, the app cannot be blamed...
> Memory analysis with Eclipse MAT shows that the class
> org.apache.cxf.common.util.CacheMap grows excessively large, will attach some
> screenshots (also tested with jVisualVM) and logs later.
> Possibly related bugs:
> Memory Leak in WSDLManagerImpl - SchemaCacheMap
> CXF-1621
> Memory leak due to literal keys in WSDLDefinition map
> CXF-1639
> performance of repeated calls to jaxws.Service.createPort is poor, jaxb
> context is created every time
> CXF-850
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
