Notes from Synapse F2F Sept. 22nd to 24th in Cupertino, CA

Thursday, September 22nd

Today is background day - going thru Celtix, X-Broker and Axis2. The presentations discussed will be posted here soon.

Presentations

Friday, September 23rd

Attendees

Discussing Features & Function

Specific Mediations We'd like to Have

Terminology

Usecases

Core Model

There is a rule: match -> mediation component approach

Suppose I want to log every message, and then run tranform mediation on the stockquote service.

rules * -> LogMediation, QoS markers http://xmethods.net/stockquote -> StockXFormMediation, QoS markers

This could be used as a "composition" model, but that is not the primary aim. Primarily compositions happen statically inside a mediation. So for example, we may provide a BPELMediation that use BPEL syntax to compose sub-mediations. Alternatively, we may provide some logical composition components (chain, branch, tee) that can be combined to compose mediations.

However, there is some differences between using the rule model and the mediation composition mode:

1. The invocation of a Mediation component by the framework is a single thread - so multiple rule matches may operate on separate threads (e.g. above, the logging may happen on a different thread to the xform).

2. The QoS markers are only at the level of a rule. So mediation writers can do what they like inside a mediation. So if a mediation wants to launch a new AxisEngine to get the security handlers invoked it can.

3. The rule matching is inherently dynamic, whereas the mediations are more statically defined and deployed. (of course a mediation is just Java so it can do as much dynamic stuff as any other Java - reflection, classloading, etc).

Rules can be added dynamically.

In fact because the system is inherently one-way, sometimes we will automatically add a new tule that matches an async response.

XML

We could write our own serialisation

<sxl>

</sxl>

or use Policy attachment

<wsp:PolicyAttachment>

</wsp:PolicyAttachment>

Another option would be to have an ASCII/LEX serialisation that transforms to/from the PolAttachment model above.

Synapse/200509F2F (last edited 2009-09-20 22:48:37 by localhost)