1. If you use JMS transport (or JMS bridging to HTTP for HTTP-based
services) you can do async web service invocations using either Axis or
WSIF. There used to be a page on using JMS with Axis in the Axis docs.
WSIF has some sample programs to examine. The WSIF version has
something like the callback you want.
2. You could also consider using adapter proxy web services to convert
the synchronous ones into async (this is more or less what JMS is doing
for you in the approach mentioned above). Operation A in service SA
might have an input a and output a'; you have a proxy Operation AA in
service SAA which has only input a and a callback (say using
WS-Addressing) to receive the result. In Operation AA would pass on the
input and return http status 200. When it got the output, it would send
it to the callback address. Your client would have to be listening for
these according to some protocol like soap over http, and would have to
have some correlation id to associate responses with the original
requests, etc.
Jeff
Tom Oinn wrote:
Hi Guenter,
Thanks for the reply. Our problem here is that we don't control the
services (this system integrates whatever it finds, that's the point)
and that when I call call.invoke(...) it blocks until there's a response.
What I'd ideally need is something that allowed me to specify a
callback of some kind to call.invoke and have the initial method call
return immediately. This would then have to be combined with some
agent that received _all_ soap messages (otherwise you just have
another thread listening per service and it defeats the point) and
invoked the callback when a message arrived that was in some way
correlated to the initial request.
I have a suspicion I'm moving into 'write your own soap toolkit'
territory again...
Tom
Grossberger, Guenter wrote:
Hi, Tom!
First I don't know if something like that is covered in Axis 2.
But I guess the Half-Sync/Half-Async pattern or an other event-handling
or concurrency pattern described in the POSA2 book (Pattern-Oriented
Software Architecture: Patterns for Concurrent and Networked Objects
Vol. 2, http://www.cs.wustl.edu/~schmidt/POSA/) would work for you.
Bye,
Guenter
-----Original Message-----
From: Tom Oinn [mailto:[EMAIL PROTECTED] Sent: Wednesday, June 08, 2005
2:19 PM
To: [email protected]
Subject: Minimising thread usage across multiple axis calls
Hi all,
Consider this problem - we have a set of services outside of our
control (a mixture of doc/lit and rpc, that's not important though)
which are being orchestrated into a dataflow style workflow. At
present we must allocate one thread to each service invocation and
this is potentially a problem for cases where either a single
workflow launches thousands of services concurrently or where the
workflow service itself is hosting a large number of active workflows.
In principle it would seem possible to have a thread which was aware
of all currently pending SOAP operations (where the request message
had been sent but no response received) and which could poll some
incoming queue regularly, launching any neccessary threads if
messages are present.
This implies a client manager component, I'm not aware of anything
analogous to this in the current architecture though?
Questions :
1) Is this covered in the Axis2 design in any way?
2) Is anyone else interested?
Cheers,
Tom
