Wx-Tx and Transactions. Don Box's Spoutlet: "ACID Transactions aren't meant to span across org boundaries... WS-Tx is... actually quite difficult to get right. [snellspace]

...a transaction and 2PC with nodes that could or could not be there. Yikes. [Sam Gentile's Radio Weblog]

Back from vacation and shooting from the hip without being up-to-date on these specs or the various weblogs of more informed experts... here's my quick 2 cents... I'm interested in your comments...

What I think matters is the legal document. We need simple meeting places where we can securely create and access documents that can not be repudiated.

The protocols for using these meeting places should not matter too much. They will evolve. There will more than one protocol in use. All kinds of devices and systems have to be able to reach these meeting places to exchange documents.

The lowest level protocols should be simple, secure, reliable exchanges of documents with meeting places. We should not try to build the perfect, or even imperfect, self-contained protocol.

Above the simple protocols should be the agreed patterns of exchange, including compensating transactions to account for breakdowns in the anticipated patterns. Possession of a document that cannot be repudiated is the key. [Patrick Logan's Radio Weblog]

Logan seems to be thinking about a deeper notion of transaction than D. Box. I'm not sure we should try to drag database terminology into a discussion of communication, negotiation, and commitment implemented with web services – or otherwise. An ACID transaction attempts to solve very different problems than a large class of potential intra-organization agreements. For example, an ACID transaction guarantees a consistent view of a database – frequently in negotiations we want to maintain in inconsistent view of an external context (perhaps prices), while the discussion moves forward.

The focus of the common database terminology is: make this true. The focus of many negotiations is more like: what do we want to become true.

Perhaps Box has it wrong. Its not that a two phase commit is hard, it’s the wrong thing for many problems. Where the semantics are appropriate, I’m not convinced that the implementation is that difficult – if I’m negotiation for 10,000 keyboards at $1.04 each, I do want to make sure I don’t spend all my money on 100 monitors at $104 each – I want consistence of view and if the price depends upon a shipping date I want both guaranteed before anything is final.