- From: Sergey Beryozkin <sergey.beryozkin@iona.com>
- Date: Tue, 7 Nov 2006 12:53:48 -0000
- To: "Yalcinalp, Umit" <umit.yalcinalp@sap.com>, <public-ws-policy@w3.org>
- Message-ID: <007601c7026b$c4d13670$3901020a@sberyoz>
Hi
Here's a use case I suggest to be considered while deciding how to close this issue..
We have what we call HA services in our product :
<service name="m1:HelloWorld">
<port name="Port1">...<port/>
<port name="Port2">...<port/>
<port name="Port3">...<port/>
</service>
There's nothing HA specific in the service definition, just multiple endpoints and we explicitly configure a client runtime to add a specific plugin which lets the client automatically enumerate between different endpoints if one of those endpoints fail. If ha plugin is not in the client config then the client still is able to talk to this service, by picking up the first port for ex.
The service runtime is configured under the hood to ensure the fault-tolerance semantics are preserved, that is it can guarantee HA all the time to all requesters which know what it is.
Now imagine WS-Policy comes along. We can say :
<service name="m1:HelloWorld">
<port name="Port1">...<port/>
<port name="Port2">...<port/>
<port name="Port3">...<port/>
<Policy>
<All>
<iona:highlyAvailable/>
<ws:secure/>
</All>
</Policy>
</service>
The client doesn't need to be configured explicitly. If it's our client runtime (or indeed a third-party one) which understands what <iona:highlyAvailable/> is then a required plugin will be loaded automatically when it sees <iona:highlyAvailable/> assertion. Zero config.
In other words, if a policy-aware client encounters iona:highlyAvailable and recognizes assertion then this means a client runtime will load a plugin which will tell the client runtime that if a communication with the service/port1 fails then the communication should continue through other available ports, service/port2.
This assertion has following properties : no wire representation and it can be ignored by unaware clients.
If it's a client runtime which has no clue what <iona:highlyAvailable/> is it will still be capable of talking to this service by selecting the first service endpoint for ex.
The important thing is that <iona:highlyAvailable/> may be of genuine interest to requesters.
Now imagine a third-party policy-aware client coming along which has this requirement :
<Policy>
<ws:secure/>
</Policy>
It tries to match our policy above and fails because it does not know what <iona:highlyAvailable/> is.
The failure is a setback for a service because it's secure as expected by a third party client and a client does not need to know what <iona:highlyAvailable/> is.
We believe the following expression is adequate :
<Policy>
<ExactlyOnce>
<iona:highlyAvailable wsp:optional="true"/>
<ws:secure/>
</ExactlyOnce>
</Policy>
So that HA-unaware client can skip it.
We're not convinced additional markers will simplify the situation from a practical point of view even though they might add some clarity from a theoretical perspective. We'll welcome a new marker if it will truly help both users and tools to distinguish between different assertions...we'd ask a group to work with this case at least when deciding on how to close the issue...
Futhermore we'd like some more clarity in what different policy alternatives mean. We feel that policy alternatives should not be considered to be related to each other and that it's immaterial from a client's point of view what other assertions a given provider might expose, a client's view of the service is through the given selected alternative only, a provider may have/do many more things, but it's a detail a client should not be concerned about...
Regards
Sergey
----- Original Message -----
From: Yalcinalp, Umit
To: Sergey Beryozkin ; public-ws-policy@w3.org
Sent: Saturday, November 04, 2006 1:06 AM
Subject: RE: NEW ISSUE :Clarify usage of assertions with no behavioral requirements on the requester
Folks,
Here is my proposal to close issues [3789] and [3721]. Call it proposal 5.
I am including here only the changes that will impact the framework. There has been quite a discussion on the use cases for both of these issues. I will be happy to draft text to illustrate the use of the attribute if we accept the proposal for specifically the guidelines document and the primer. Since we are trying to close all the issues on the framework, I targeted text for this section only. Note that for the configuration parameters, only the providers tools will understand them. For those which have no behavioral requirements and not affect the wire, the client can choose to understand the vocabulary.
IMO, this proposal has the advantage of addressing different uses of an assertion, allows different consumers to be able to include them into their intersection algorithm, allows the provider only or client tools to process intersections in a uniform way and thus satisfies both the issues at hand with a minimum disruption to the framework.
If you think we should include a "default" value to the attribute and assume all assertions exhibit a predetermined value to target all consumers, I am open to that. However, the way I wrote it, the assertions that do not have this attribute do get included into the alternative vocabulary so, that would be an addition which may not be necessary.
Note that I am not tied to the name of the attribute. If you want to call it mustUnderstand, I do not care, but I like my name better. These assertions are included only when the consumer understands them.
[3789] http://www.w3.org/Bugs/Public/show_bug.cgi?id=3789
[3721] http://www.w3.org/Bugs/Public/show_bug.cgi?id=3721
--------------------------------
Here is the proposal:
Section 3.1: Add the following section
3.1.1 wsp:include attribute information item
WS-Policy framework provides a global attribute information item to enable authors to selectively designate the consumers of the assertion based on the value of the attribute.
A [local name] of include .
A [namespace name] of "http://www.w3.org/@@@".
The type of the include attribute information item is a list xs:anyURI. Its actual value MUST be an absolute IRI as defined in [IETF RFC 3987].
The presence of this attribute indicates that the assertion is targeted to a specific set of consumers for processing this attribute. Although the type of the assertion is part of a policy expression's vocabulary it appears, however it is only utilized by targeted consumers for intersection [See Section 4.4 Policy Intersection]. Authors should not utilize this attribute for assertions that are required for interoperability targeted to all consumers.
Changes to Section 4.4
Replace the paragraph:
{As a first approximation, an algorithm is defined herein that approximates compatibility in a domain-independent manner; specifically, for two policy alternatives to be compatible, they must at least have the same policy alternative vocabulary (see Section 3.2 Policy Alternative).}
With the following/Make this sentence a paragraph:
{As a first approximation, an algorithm is defined herein that approximates compatibility in a domain-independent manner; specifically, for two policy alternatives to be compatible, they must at least have the same policy alternative vocabulary (see Section 3.2 Policy Alternative) subject to the assertions that are present in an alternative that are marked with wsp:include attribute in the vocabulary. Each assertion that uses wsp:include attribute is included in the alternative only if a consumer of the alternatives (for example, a requestor) handling the intersection understands the value of the wsp:include URI value. Otherwise, the assertion type is not considered part of the alternative. The following rules apply after this processing: }
----------------------
Dr. Umit Yalcinalp
Architect
NetWeaver Industry Standards
SAP Labs, LLC
Email: umit.yalcinalp@sap.com Tel: (650) 320-3095
SDN: https://www.sdn.sap.com/irj/sdn/weblogs?blog=/pub/u/36238
--------
"First they ignore you, then they ridicule you, then they fight you, then you win." Gandhi
"You die. Gandhi was an optimist" Roberto Chinnici
----------------------------------------------------------------------------
From: public-ws-policy-request@w3.org [mailto:public-ws-policy-request@w3.org] On Behalf Of Sergey Beryozkin
Sent: Wednesday, Nov 01, 2006 10:42 AM
To: Sergey Beryozkin; public-ws-policy@w3.org
Subject: Re: NEW ISSUE :Clarify usage of assertions with no behavioral requirements on the requester
Hello
I'd just like to explore a Proposal3 below in a bit more detail, copying it here for convenience :
Proposal 3. Introduce a new attribute wsp:provider-only and leave wsp:optional the way it is now.
Ex :
<Policy>
<m:MTOM wsp:optional="true"/>
<sp:security/>
<!-- of potential interest to requesters -->
<m:highlyAvailable wsp:provider-only="true"/>
<!-- server-specific stuff, of no interst to requesters-->
<m:myCustomServerLogging wsp:provider-only="true"/>
</Policy>
I'd like to remove server-only private stuff from here and concentrate only on capabilities which can be of interest to clients. Additionally, just for the purpose of this example I'll use wsp:advisory instead of wsp:provider-only because I just like wsp:advisory the name :-). Lets assume for a sec wsp:advisory is good for advertizing capabilities which can be ignored, server-specific stuff does not go here at all :
<Policy>
<m:MTOM wsp:optional="true"/>
<sp:security/>
<!-- of potential interest to requesters -->
<m:replicatable wsp:advisory="true"/>
</Policy>
Normal form :
<Policy>
<ExactlyOnce>
<All>
<m:MTOM/>
<sp:security/>
<m:replicatable wsp:advisory="true"/>
</All>
<All>
<sp:security/>
<m:replicatable wsp:advisory="true"/>
</All>
</ExactlyOnce>
</Policy>
With wsp:optional we only mark optional assertions with some behaviours a requester will need to do.
With wsp:advisory we only mark things a requester may choose either to notice or ignore. Note that in the case of
I've thought more about it. I feel that it's kind of nice to have this separation of concerns. It makes sense. However I'm not sure that it makes things simplier/less unambiguous and confusing from a practical perspective :
* there's some overlap here in that both <m:MTOM/> and <m:replicatable/> are assertions a requester may optionally ignore. I just feel it will cause a lot a lot of confusion for users to figure out what should be marked as wsp:optional and wsp:replicatable (and then withsome standard wsp:local attribute marking pure server-only assertions :-)). I think the above example shows there's some ambiguity in how <m:replicatable wsp:advisory="true"/> should be interpreted, it is in the alternative, but at the same time it's not because it can be ignored, it kind of dilutes the meaning of alternative.
* people would still be able to make <m:replicatable/> with wsp:optional
* people would likely start marking server-only assertions like <m:myCustomServerLogging/> with wsp:advisory/as well as with wsp:optional.
* it's not clear what a requester seeing <m:replicatable wsp:advisory="true"/> should do if does not understand what it means : just ignore it, or ignore it and warn a user ?
* it's not clear what a requester seeing <m:myCustomServerLogging wsp:advisory="true"/> should do if does not understand what it means : just ignore it, or ignore it and warn a user ? this is an example of misused <m:myCustomServerLogging/>
* I think it will slow down the requester's processing because even though <m:replicatable wsp:advisory="true"/> is marked as something which can be ignored a requester will still to process it and check if it's something it actually understands or not...
So far Proposal1 with a possible renaming to wsp:ignorable (and controversial 2 :-)) or Proposal4 (workaround) work for me...they just seem simplier, don't really require to do anything major. I'm not sure adding a new attribute will simplify/clarify things but lets think more Proposal3 too...
And I can't help waiting reading a Proposal 5 from Umit :-)
Enjoy
Sergey
Hello,
As part of the action assigned to me at the yesterday's concall, I'd like to offer to your attention 4 alternative proposals on how to resolve this issue which I believe have been mentioned before. I hope the subsequent discussions will point to a best/preferred/least controversial/simpliest/easy to understand solution.
I'd like to clarify that the purpose of resolving this issue is to have a guideline to policy authors which wish to advertize some of provider's capabilities. For ex, custom:free, custom:infoConfidential, custom:highlyAvailable, custom:replicatable which requesters can ignore or do something about.
I'll add my own comments (S.B) to each of the proposal. However, please do not consider them as something which represents the position of Iona at this stage.
Proposal1.
Drastically simplify the meaning of wsp:optional. Explain that wsp:optional marks an assertion which can be ignored by a requester. In other words wsp:optional is identical in meaning to wsp:ignorable. Clearly state that wsp:optional assertions are by no means optional to a provider.
Explain that wsp:optional is a shortcut which simplifies creating different policy alternatives/vocabularies.
Ex : <custom:infoConfidential wsp:optional="true"/> means a requester can ignore it.
<S.B> IMHO this is the simpliest solution which works. IMHO the current treatment of wsp:optional is too complicated. For ex, I think stating that if there's something a provider always does should not be marked as optional, otherwise it has to be optional will only confuse the users. IMHO it's wrong to say a provider optionally does MTOM, it always does it, one requester can choose an alternative with no MTOM but it does not mean the provider does not do it with the other requester. In other words an alternative is a piece of vocabulary. If MTOM is not in the selected alternative or not it does not mean that a provider has lost its capability to do MTOM. Viewing wsp:optional as a simple marker to indicate ignorable assertions is a very simple and working solution IMHO.
<S.B> additional advantage is that it cleanly alligns with a proposed wsp:local attribute in that a user will be guided to mark server-specific stuff as being wsp:local. Imagine a GUI asking a question :
* "Is this assertion must be understood by a requester", YES-normal assertion.
* "Is this assertion may be ignored by a requester", YES-multiple vocabularies are created, addional question : "Can this ignorable assertion be of any interest to a requester ?" NO - mark it as wsp:local.
Nothing will prevent a user by exposing <myLocalServerOnlyAssertion/> by marking it as wsp:optional. This approach will proactively teach a user not to do it and only expose assertions which can be of interest to a requester.
Proposal 2. Similar to proposal1. Drop wsp:optional altogether and find out how simple things have become. The reason it works is that if we forget about wsp:optional for a second, we can easily see that if one assertion is contained in one alternative and not in the other one then it's an optional/ignorable assertion. Another reason it works is that we can imagine policy authors using GUI tools which guide them. Imagine questions like : "Is this assertion must be understood by a requester" ? "Is this assertion may be ignored by a requester" ? Yes to the last question will result in a tool creating two alternative vocabularies.
<S.B> the same comments as above
<S.B> disadvantage is that more work will be requiored in a manual edit mode.
Proposal 3. Introduce a new attribute wsp:provider-only and leave wsp:optional the way it is now.
Ex :
<Policy>
<m:MTOM wsp:optional="true"/>
<sp:security/>
<!-- of potential interest to requesters -->
<m:highlyAvailable wsp:provider-only="true"/>
<!-- server-specific stuff, of no interst to requesters-->
<m:myCustomServerLogging wsp:provider-only="true"/>
</Policy>
<S.B.> This works only if wsp:provider-only="true" are not stripped but this means wsp:provider-only="true" pointing to server-specific stuff only won't be stripped and be consistently leaked (and confuse a user at a design-time). If it's stripped then we'll lose <m:highlyAvailable/> which is of interest to knowledgeable requesters. IMHO it's a can of worms.
Proposal 4. Noop. Think of workarounds : multiple endpoints, different Policies, WSDLs, etc...
Enjoy, Sergey.
Received on Tuesday, 7 November 2006 12:53:12 UTC