A significant problem in the trenches for Web Services isn’t the ability to play nice or not with existing Internet protocols, or its complicated type system, or Object/XML impedence. It’s optionality. When you get down to it, too much in Web Services is optional. Calling the specs composable and thus dressing that up as feature is pointless - optionality is a curse. Yes, WS specs are designed to be composable, but this underplays the fragmentation costs of not having an interoperable core.
Optionality here means: a) the state space for communications is huge , b) everything has to be negotiated, c) no clear layering resulting in dependency issues. These introduce complexity and thus drive up costs. As a result I disagree that implementing something like WS-Eventing is “simple”, exactly because there is freeform composition. And that’s just WS-Eventing. What about addressing, transfer, policy, wsdl, faulting, the various reliable messaging approaches? The list goes on and on. Even the basic SOAP stack seems to be difficult to get right; we’re on the third generation inside 7 years of what is supposed to be stable computing infrastructure.
Favouring optionality can’t possibly be a good idea for protocols, or even protocol frameworks like SOAP. In that sense RSS or Atom have more interop potential, precisely because they take options away by design, even though they wouldn’t present an obvious choice for mainstream enterprise purposes.
Here’s Elliotte Harold and Ted Neward arguing over the merits of HTTP as a Highlander Protocol.
“Web Services Addressing is trying to define an addressing scheme that can work over HTTP, SMTP, FTP, and any other protocol you can imagine. However, each of these protocols already have their own addressing systems. Developers working with these protocols don’t want and don’t need a different addressing system that’s marginally more compatible with some protocol they’re not using in exchange for substantially less compatibility with the protocol they are using. Besides nobody’s actually doing web services over anything except HTTP anyway. Doesn’t it just make more sense to use the well understood, already implemented debugged HTTP architecture for this instead of inventing something new? “
“No, it doesn’t. There’s no sense in trying to take bits that were designed for a distributed hypermedia system (Fielding’s words, not mine) and trying to bend it to fit a problem space that isn’t distributed hypermedia. Can we learn from the REST architectural style? Absolutely–and the new WS-* specs, including SOAP, do exactly that, favoring self-descriptive messages over RPC calls as the principal abstraction to deal with. But does that mean we tie ourselves solely to HTTP? Oh, hell no.”
The problem with using HTTP as the Highlander protocol has been described best by Marshall Rose:
"HTTP has become the reuse platform of choice, largely because: - it is familiar - it is ubiquitous - it has a simple request/response - it usually works through firewalls These are all good reasons, and - if HTTP meets your communications requirements - you should use it. The problem is that widespread availability of HTTP has become an excuse for not bothering to understand what the requirements really are. It's easier to use HTTP, even if it's not a good fit, that to understand your requirements and design a protocol that does what you really need."
HTTP bias aside, sometimes you do have to work across multiple protocols. That results in the problem of managing and preserving addresses and message identity. And when you do have to span protocols and systems for messaging purposes, especially where there are multiple self-interested administrations, you want to have a standard envelope that provides at minimum a means of identifying the message, the sender, the recipients, and probably time stamping. Some kind of ability to correlate messages. These are baseline criteria - without them you’re toast. You need standard headers not just for routing traffic (important) but because you need to know what the heck is going on operationally - you want a standard means of asking the various endpoints (or nodes, or gateways, or hubs, or whatever you want to call them) questions. Where is my message? How many messages did you drop? Which messages did you drop? When? Whose messages are you dropping? Will you resend this message? How many messages have you sent to so and so? What open items do you have? And so on. However WS, and in particular, SOAP, doesn’t provide a standard (as in non-optional) set of headers and thus it doesn’t provision this kind of generic communication facility. I’ll confess that this has never made sense to me, given SOAP’s protocol framework and inter-systems ambitions.
Ted has a specific rant about broadcast and push being unsuitable to run over HTTP. I don’t think there’s any argument that, formally speaking, HTTP is not the right protocol for those jobs. But here’s the kicker. Blogs and RSS syndication are exactly that, done over HTTP infrastructure. The world has not gone running to push RSS over XMPP because there is obviously a scale problem when you draw things out on a napkin. [Which is not to say this writer wouldn’t like to see more Atom over XMPP; it’s clearly a good idea. Nonetheless the Web sky is not falling in yet. ]
And even in the push scenario, what’s the point of a protocol framework? Why not use XMPP for push/pubsub? Does this mean when people on the ground start getting Eventing systems and EDAs built out over commodity IM, can we expect to see another rounds of the “It’s Not Sufficient” meme played out again, this time for messaging protocols?
What now for SOAP?
If you want to span systems asynchronously in a manageable fashion you’ve got to have standard models for addressing and identity baked in. These are not optional if this stuff is expected to work. It’s taken some time for the WS community to get round to dealing with addressing and naming, but this is key stuff for getting something done over a gateway and for me, SOAP’s value is largely in dealing with gateways asynchronously. Addressing being an optional add-on to SOAP is quite strange then, although now that it exists that’s good news. Down the line then what would be useful is a subset and hardening of SOAP; a 1.3 revision where WS-A and some key message headers are non-optional.
WS should be starting out with, and not ending with, profiles like RAMP.