Hi David,
For SOAP, a standard CXF client uses a CXF class named HTTPConduit,
which internally uses the java.net.HttpURLConnection class.
For the latter, a JSE class then, you can set 2 timeouts,
- the socket connect timeout
- the socket read timeout (since 1.5)
CXF's HTTPConduit allows clients to set these values through so-called
policies
for which there are (resp) two getters/setters:
get/setConnectionTimeout
get/setReceiveTimeout
In addition there is an "overall timeout" called Synchronous Timeout.
You can all change these values whenever you want (i.e. between calls) in a
way similar to this :
private static final QName SERVICE_NAME = new QName("
http://my.service.com/";, "MyService");
private static MyService port ;
private HTTPConduit conduit ;
private HTTPClientPolicy policy ;
(...)
URL wsdlURL = MyService.WSDL_LOCATION;
MyService ss = new MyService(wsdlURL, SERVICE_NAME);
port = ss.getMyServicesSoap();
// Get client
Client client = ClientProxy.getClient( port );
// Set response overall timeout to 5 minutes
((ClientImpl) client ).setSynchronousTimeout( 5 * 60 * 1000);
// Set Connection and socket timeout to 2 minutes
conduit = (HTTPConduit)client.getConduit();
policy = conduit.getClient();
policy.setConnectionTimeout ( 120000 );
policy.setReceiveTimeout( 12000 ) ;
(...)
Where MyService and MyServiceSoap are classes generated by jaxws wsdl2java
(e.g. from CXF).
Does that fit your requirements ?
Alain Pannetier
On Tue, Jul 6, 2010 at 11:30 PM, KARR, DAVID (ATTSI) <[email protected]> wrote:
> I work with some middleware that makes SOAP calls with an ad hoc
> internal framework using HttpClient 3.0.1. We set the "socket timeout"
> on HttpClient, but we were really intending that to be an absolute
> timeout on the web service call. The socket timeout is really only used
> "per packet" so it's very easy for absolute times on service calls to be
> well over the "read timeout" value.
>
> As I don't think I can fix this with HttpClient out of the box, I'm
> investigating the possibility of having a background task kill
> connections if they've gone past their time limit.
>
> What I'd like to ask here, though, is whether CXF has any direct or
> indirect support for solving this problem. I've been considering
> putting CXF in here instead of the ad hoc framework combined with raw
> HttpClient, but I'm looking for specific points to support this. If I
> could more easily support this "absolute time out" feature, that would
> be a plus.
>
