Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Fixed
-
2.2.7
-
Standalone CXF-client used on Windows Vista
Description
Hi,
I'm trying to invoke 10000 external services using invokeAsync from my standalone JAX-WS client.
I set my custom Executor on the Service objects. This executor is a ThreadPool with 3 threads. But it looks like it is ignored completely by the CXF
run-time. Instead of an Executor, an instance of AutomaticWorkQueueImpl is used.
BTW, by default, AutomaticWorkQueueImpl is an unbounded queue, so that a few thousands threads are created for processing asynchronous responses. This is of cause problematic and has a big performance impact. By providing the configuration for AutomaticWorkQueueImpl in cxf.xml, the AutomaticWorkQueueImpl can be configured to have a required number of threads and required maximum capacity.
So, I did some debugging to see, why and where AutomaticWorkQueueImpl is called at all during asynchronous WS invocations using Dispatch.invokeAsync().
This is the stack trace I got:
Thread [main] (Suspended (entry into method execute in AutomaticWorkQueueImpl))
AutomaticWorkQueueImpl.execute(Runnable) line: 247
HTTPConduit$WrappedOutputStream.handleResponse() line: 2153
HTTPConduit$WrappedOutputStream.close() line: 1988
HTTPConduit(AbstractConduit).close(Message) line: 66
HTTPConduit.close(Message) line: 639
MessageSenderInterceptor$MessageSenderEndingInterceptor.handleMessage(Message) line: 62
PhaseInterceptorChain.doIntercept(Message) line: 243
ClientImpl.invoke(ClientCallback, BindingOperationInfo, Object...) line: 422
ClientImpl.invokeWrapped(ClientCallback, QName, Object...) line: 371
DispatchImpl<T>.invokeAsync(T, AsyncHandler<T>) line: 288
TestAsyncProviderClient.invokeMyDispatch(Dispatch<Source>, Object, AsyncHandler<Source>) line: 298
TestAsyncProviderClient.testManyAsyncResponses() line: 218
TestAsyncProviderClient.main(String[]) line: 159
According to this trace, WS-response processing at the HTTP level is put for the execution on a dedicated working queue in the HTTPConduit.handleResponse method:
protected void handleResponse() throws IOException {
handleRetransmits();
if (outMessage == null || outMessage.getExchange() == null || outMessage.getExchange().isSynchronous())
{ handleResponseInternal(); } else {
Runnable runnable = new Runnable() {
public void run() {
try
catch (Exception e)
{ Message inMessage = new MessageImpl(); inMessage.setExchange(outMessage.getExchange()); inMessage.setContent(Exception.class, e); incomingObserver.onMessage(inMessage); }
}
};
WorkQueueManager mgr = outMessage.getExchange().get(Bus.class)
.getExtension(WorkQueueManager.class);
AutomaticWorkQueue queue = mgr.getNamedWorkQueue("http-conduit");
if (queue == null)
queue.execute(runnable);
}
}
It is easy to see that Executor set for the JAX-WS Service is not propagated down to the HTTP transoport level, which required dedicated configuration for AutomaticWorkQueue, be it http-conduit queue or automatic work queue.
For WS-responses at the user-level, i.e. by means of the JAX-WS AsyncHandler handlers, the proper Executor is taken from the Service object.
See also the discussion on the mailing list:
http://old.nabble.com/RejectedExecutionException-when-doing-many-invokeAsync-invocations-ts28266640.html