Re: [wsbpel] Removing "compensating a process as a whole"

[Assaf Arkin <arkin@intalio.com>]

Satish Thatte wrote:

>The problem for BPEL is that it either goes into monolithic "I specify
>everything" mode which we have all seen fail in the past, or it sticks
>to its factored role and takes dependencies on things like WSDL,
>admittedly moving targets, that will define the aspects it leaves
>unspecified.  If we make interoperability dependent on platform-level
>deployment modes I would argue that we have failed to maintain the loose
>coupling implicit in the web services architecture.
>  
>
I perfectly agree. I don't think we should try and boil the ocean, but 
we do need at least to interoperate.

>As for interop with WS-TX, specifically with the BA protocol(s), my
>personal opinion is that this is something we should look at very
>closely and do whatever is needed to make it happen.
>  
>
Same here.

But again we need to formalize it. It's perfectly acceptable for the 
spec to say that compensation for the process is invoked in some 
implementation specific manner. WS-TX is one such option. It's also 
perfectly acceptable for the spec to have some non-normative reference 
to WS-TX to illustrate how one would use WS-TX if they choose to, 
without precluding using other mechanisms or not using any mechanism at all.

But imagine that two systems decide to use WS-TX and WS-TX provides all 
the information they need to determine how to send coordination messages 
around. And they also decide to use BPEL, and BPEL provides the 
definition of the business protocol which indicates the abillity to 
compensate. Are they able to interoperate?

Or is there some minimal requirement from the spec without which 
interoperability is not possible? If the spec doesn't at least 
investigate that minimum requirement, then interoperability would not be 
possible. To do that the spec needs to at least acknowledge that WS-TX 
or a similar protocol is one option for performing compensation.

When you do that certain issues arise which needs to be clarified, non 
of which requires a wholesale rewrite of the specification. But we can't 
resolve these issues unless we make a point that we would address 
interoperability when BPEL is used in combination with WS-TX. 
Incidentally, doing so would clarify one of the most common uses of a 
compensation handler at the process level, which is not apparent from 
the current reading of the spec.

Again, what we both have in mind may be well and good and actually work, 
but if it's not written down then it won't help interoperability on a 
large scale.

arkin

>Satish
>
>  
>