Re: EPP Poll 'msgid' attribute
Andrew Sullivan <[email protected]>
| Newsgroups | gmane.ietf.provreg |
|---|---|
| Message-ID | <[email protected]> |
On Mon, Oct 31, 2011 at 11:33:19AM +0200, Mike O'Connell wrote: > > We're looking at implementing a selectable poll message process where the op='req' can specify the clTRID as the 'msgid' attribute to retrieve any unacked responses associated with that client identifier. > > Has anyone implemented anything like this? What are the thoughts regarding this? I argued for altering the way poll worked to do something quite like this, about the time that EPP went through one of the standards levels (I think from proposed to draft). I outlined the suggestion in a long-expired I-D that is still available: http://tools.ietf.org/id/draft-sullivan-epp-experience-00.txt. See sections 3.4 and 4.2. I strongly recommend you not try to shoehorn this functionality into the existing poll queue mechanism. The existing mechanism is quite clearly FIFO, and if you try to manage it differently while using the same syntax, you not only produce quite different behaviour with the same protocol; you'll also find that you don't have enough primitives to manage the queue correctly, and messages will get lost. Anyway, you'll be in flagrant violation of section 2.9.2.3 of RFC 5730: "a successful response to a <poll> command MUST return the first message from the message queue." If conformance is important to you for contractual reasons, then you just can't do this. When I argued in favour of the approach to the poll queue I outlined in that I-D, Scott Hollenbeck pointed out (correctly) that you could do it with an extension anyway. That's what I'd recommend. Afilias never did undertake that work, however, because a colleague of mine in the development department decided instead that the passage, "Service messages SHOULD be created for passive clients affected by an action on an object," used the ordinary English meaning of SHOULD and not the RFC 2119 one. (Note that in RFC 3730, that SHOULD was a MUST, but I still don't see how the SHOULD gets one off the hook in case it's merely convenient.) Best, A -- Andrew Sullivan [email protected] _______________________________________________ provreg mailing list [email protected] https://www.ietf.org/mailman/listinfo/provreg