synapse-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From ant elder <>
Subject Re: ClassMediator vs. mediators as Axis2 services
Date Fri, 06 Jan 2006 13:09:11 GMT
The benefit is not forcing everyone to nearly always put
in their synapse.xml

The requirements are:

1) Some mediators require addressing headers, some don't
2) Some mediators require addressing headers but aren't invoked through an
3) Don't want addressing module to run multiple times due to overwriting
altered headers
4) Sometimes addressing headers aren't required at all so don't want
addressing module run (for performance?)

So the current solution  is:

A) Require explicit addressing when and where required in synapse.xml

But there are other solutions:

B) Synapse by default always calls addressing at start for every message.
Optional <dont-engage-addressing/> for (4)

C) Have a way for mediators to indicate they require addressing so Synapse
can engage the module when required. And Synapse knows if it was already
engaged for a previous mediator so only engages it once. (Or have the Axis2
addressing code itself see its already run so doesn't need to again)

Maybe I don't understand the Axis2 module stuff properly, but is there also:

D) Iff a Mediator calls one of the WSA getter's/setters on SynapseMessage
then if it hasn't already the addressing code gets run to populate the

Surely (4) isn't so common, all of Synapse samples use addressing. So isn't
(B) more appropriate than (A) for now, and look at  (C) or (D) for later?


On 1/6/06, Paul Fremantle <> wrote:
> Ant
> I don't agree that synapse should always do addressing. Firstly, if we
> have addressing engaged and we make multiple passes through Axis then it
> overwrites the To/From/etc properties in the SynapseMessage every time.
> Secondly, I don't agree that we always want to parse those headers. In
> fact we decided at the very first F2F not to.
> We started out making every mediator a pass through Axis2 for a specific
> benefit - QoS. However, we now have an alternative model to get those QoSs
> (the emptymessagereceiver) and I don't see the benefit. I think there is a
> clearer model between Axis2 and Synapse if we only call back into Axis2 when
> we need a specific Axis2 service.
> Paul
> On 1/6/06, ant elder <> wrote:
> >
> > Is the ClassMediator really needed?
> >
> > In yesterdays IRC chat there was some discussion about mediators being
> > Axis2 services vs. the ClassMediator which calls a Java class directly
> > without going through an Axis engine. It was said the ClassMediator is
> > easier to use as you don't need a service xml and is faster.
> >
> > Are these the only reasons for having the ClassMediator? Wouldn't having
> > Synapse auto-deploy to Axis2 (as Glen suggested) fix the complexity problem,
> > and if it really is so slow going through an empty AxisEngine shouldn't we
> > try to get the Axis2 guys to fix that instead of us just not using Axis?
> > Surely a pass through an empty engine should just be a few empty loops and a
> > bunch of if statements which shouldn't add so much overhead?
> >
> > One thing this would help with is with the addressing stuff. i don't
> > really like having <engage-addressing-in/> explicitly in the Synapse XML.
> > Shouldn't Synapse just know when addressing is required? If you have a
> > Mediator the routes based on a WSA header then if the mediator was an Axis2
> > service the addressing stuff could be engaged automatically on the pass
> > through the engine.
> >
> >    ...ant
> >
> --
> Paul Fremantle
> VP/Technology, WSO2 and OASIS WS-RX TC Co-chair
> "Oxygenating the Web Service Platform",

View raw message