Re: [ws-rx] An alternative approach to make anon reply-to and sync rmwork

[Tom Rutt <tom@coastin.com>]

Paul Fremantle wrote:
Just one clarification.

It seems that you are considering each http request starts is own tcp 
connection.  If this is the
case in your scenario, my comment on http piggybacking is not relevant.

> Tom
>
> Answers inline.
>
> Tom Rutt wrote:
>
>> I assume we are talking about a wsdl resquest/response operation, 
>> which is using ws-rm for both the request and response, the offer is 
>> used in
>> the create sequence from the reqest sender to the request receiver, 
>> and the sequence from request sender to request receiver is using the
>> anon acks to for the request.
>
> Yes. At least some of the operations involved are two-way. Of course 
> some may have alternative MEPs.
>
>> Now it is not stated, but I assume the acks to for the offer sequence 
>> must be an asych callback (i.e., not anonymous).
>
> Yes. The acksTo for the offered sequence is a real address (e.g. the 
> same address as the server).
>
>> I interpret this case to state that the rmd has to respond to 
>> pigybaccked http requests in the same order they are received.  
>
> I'm not sure I understand this sentence.
>
>> Now
>> I assume that anon acks to means that the ack header can apear on any 
>> http response which has a request carrying that sequence number in a
>> sequence or ackrequested header.
>
> Yes I agree.
>
>> Also, I cannot see a useful scenario for sending the ack before 
>> sending the response for a given request message.
>
> There is no useful scenario. However, if the client sent a request-two 
> before it had got response-1, it may get an ack for request-2 but no 
> ack for request-1.
>
>> In face I see a case for having the ack send only after multiple 
>> response are sent, on any one of the responses for that sequence.
>
> I think the server should ack whatever messages its RMD has received 
> on any response to that sequence. But I don't see the ack behaviour 
> should be different to an ordinary sequence.
>
>> I thought the original request sender could send an ack requested 
>> without a message body or sequence header, thus there would be no
>> message number for the sequence.  Thus the request sender can request 
>> acks at any time.
>
> I agree the client can request acks any time.
>
>>
>> Tom Rutt
>>
>


-- 
----------------------------------------------------
Tom Rutt	email: tom@coastin.com; trutt@us.fujitsu.com
Tel: +1 732 801 5744          Fax: +1 732 774 5133