Hi Savas, "Use a good idea again instead of generalizing it. A specialized implementation of the idea may be much more effective than a general one." (Lampson, "Hints for Computer System Design") or as he put it more bluntly in the same paper - "Don't generalize; generalizations are generally wrong."
You are assuming HTTP here while most (if not all) of WS-* specs do not make that assumption. While what you are describing with HTTP headers may work indeed, they are all tied to HTTP.
WS-I Basic Profile does mandate HTTP, though it may not be regarded as a WS-* spec. ATOM is another example of a spec that does assume HTTP - perhaps specs that do assume HTTP don't want the WS- association. I recognise that HTTP is not the solution to every distributed computing problem but I think it fits this problem well. I do think SOAP services and RESTful services can work together, particularly in this case were the lower level protocol can be used to bootstrap the higher level protocol.
We are now seeing SOAP processors supporting other transport protocols as well (e.g. Indigo and Axis 2 support SOAP over TCP/IP). (note that I am treating HTTP as a transport protocol here and now as an application protocol :-)
When can we expect a WS-I profile that includes a transport protocol other than HTTP? What would be the first one? The solution I suggested would mean that all clients that want to resolve "handles" must support the HTTP protocol - this could be seen as a restriction but if you look at the alternative.... The cost of supporting other transport protocols. *) Now a client has to support mutliple transport protocols, increasing client complexity - a resolver service provider is free to chose whatever protocol he wants, clients have to handle it. At a minium the client will have to support HTTP as it will be the most common transport protocol. (So forcing clients to support HTTP in the HTTP only case above is not such a big deal) *) Given a Resolver EPR that uses TCP embeded inside another EPR how do I use it. I open a connection send the message, do I wait for a response out of the same connection or do I close the connection and wait for the server to respond to me asychronously. HTTP has a well defined and understand request/response pattern. For some tranpsort protocols you will need policy to declare how to use them. *) You have spec up your faults equivalent to HTTP 404 (EPR not found), 410 (EPR gone) etc. You still have to deal with HTTP 404 etc. if you use HTTP as the underlying transport protocol... *) You have to spec up and implement the caching and last-modified features provided by HTTP if you want the nice support they provide. *) Some transport protocols may not support TLS, given a resolver EPR embeded in another EPR - how does a client know the security policy of the resolver service, (eg what should be signed, encrypted, username/passwd or X509 etc...) It seems to me the complexity is getting out of hand - the WS-* answer is to build profiles, profiles which mandate transport protocols etc... cheers Mark
Regards, -- Savas Parastatidis http://savas.parastatidis.name
-----Original Message----- From: owner-ogsa-wg@ggf.org [mailto:owner-ogsa-wg@ggf.org] On Behalf Of Mark McKeown Sent: Friday, June 10, 2005 4:32 PM To: Mark Morgan Cc: ogsa-wg@ggf.org Subject: RE: Resilient References? (Re: [ogsa-wg] ogsa London f2f minutes uploaded)
Hi folks, I have read the WS-Naming strawman and been trying to follow the discussion on resilient references on the list. Putting EPRs inside EPRs seems a bit strange to me - the following seems simpler (the obvious inspiration is tinyurl) ...
<wsa:EndpointReference> <wsa:Address>mailto:mark.mckeown@manchester.ac.uk</wsa:Address> <ws-name:Handle>http://www.thegrid.com/a4dfg</ws-name:Handle> </wsa:EndpointReference>
Where "Handle" is used as an identifier/name and is mandated to be a HTTP/S URI. To resolve the Handle a client does a HTTP GET on the URI to get a new EPR, (client gets 404 if resolution is not supported, 410 if the service/resource is gone/terminated).
I can see a number of advantages to this:
*) It is simple and easy to implement.
*) The Cache-Control headers in HTTP can be used to pass on information to the client to indicate how long it can cache an EPR for - this is an optional optimization.
*) If-Modified-Since HTTP header could be supported so a client will only retieve a new EPR if the one it has is stale - again an optional optimization.
*) The "Handle" is easy to pass around between users - it can be pasted into a mail message and the reciever can use HTTP GET to retrieve the EPR.
*) HTTPS can be used when trust is important.
*) The service developer can make his EPRs "resilient" whenever he wants, ie he creates an EPR with a "Handle" but only later makes the "Handle" URI live.
Effectively the low level HTTP protocol could be used to bootstrap the higher level SOAP protocol. Perhaps there are strong reasons why this approach was not considered?
cheers Mark
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Mark Mc Keown RSS Mark.McKeown@man.ac.uk Manchester Computing +44 161 275 0601 University of Manchester ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
On Fri, 10 Jun 2005, Mark Morgan wrote:
when Mark writes: "...the general idea we have is to in fact possibly have a method who's signature is EPR resolve( [EPR] ) Where the parameter EPR is optional..."
Does this "optional" mean that the schema for the input message parameter of the resolve operation will have an optional "EPR"; it should either be empty or consist of a single EPR?
Borrowing the pseudo syntax used in the "Web Services Base Notification 1.2 Working Draft 03, 21 June 2004" and replacing the given template for the NotificationMessage syntax with appropriate naming elements, what I mean is this:
<ogsa-naming:Resolve> <ogsa-naming:ResolutionMessage> <ogsa-naming:TargetEPRHint>? wsa:EndpointReference </ogsa-naming:TargetEPRHint> </ogsa-naming:ResolutionMessage> </ogsa-naming:Resolve>
If so, what does the "optional" mean for the implementors of that porttype/operation "EPR resolve( [EPR] )"? Does the service implementation have to cater for both options?
No, it doesn'. It can implement the empty parameter one, or both the empty parameter one and the one that takes an EPR as a parameter. See below for explanation.
Or can it just return an error for the one it doesn't implement?
No, this would not be correct I don't believe.
And then what does the "optional" mean for the client's processing? If the schema indicates that the client could either call the resolve operation with or without the original EPR, how does it know which option to choose? Should it just try the one it likes?
The important points to recall here are: 1) There are two EPRs involved in this communication conversation. One is the obvious one listed as an optional parameter. For clarity, let's call that parameter the "Hint EPR". The other EPR is the implicit one which would be indicated in the SOAP headers to which the communication is taking place. Let's call that EPR the "Resolver EPR". 2) The Resolver EPR is created with whatever information the implementor of the resolution services feels is appropriate. It's up to that service to generate an EPR with enough information to do the resolution. 3) The Hint EPR is just that -- a hint. It should not be considered the end all source of information for resolution.
What we imply by all of this is that a resolution service should be able to deal with a resolution message that does not include any EPR as a parameter by virtue of the fact that the Resolution EPR to which the resolution messages are addressed is created by that service with enough information to completely carry out the resolution. If an EPR is given as a parameter to that resolution service, the service MAY choose to use that hint to more efficiently resolve the endpoint in question, but is not required to do so. In otherwords, a resolution service implementor may choose to implement code that uses the optional EPR parameter if available, but MAY also choose to ignore it.
From the client perspective, it should be the case that the client may equally choose to send either an empty message (since this is guaranteed to work), or one with an EPR (since a service may choose to ignore this parameter). In this way the client is free to pick either one -- the parameterless one if it feels inclined to go with no hints, or the message with an EPR if the client feels like this extra hint may be useful.
So, in what cases would the EPR parameter be useful should the client decide to send it and should the service decide to use it? Well, in the general case, we have found that it is tremendously useful (bordering on necessary) sometimes for a client to indicate to a resolution service that that client has already tried to communicate with a given binding (EPR) and that the resolution service might want to go to extra efforts (i.e., instead of just looking the binding up in a cache or table) to determine whether or not a more suitable EPR exists.
-Mark
Thanks, Frank.