[OpenID] Multiple endpoints in a single XRDS document

[James Tindall]

Yes, a redirect does not solve the problem of an offline OP. I think 
you're right 'pinging' the OP prior to authenticating to ensure that 
it's online may be the only solution.
It just seems unfortunate to have to add yet another RP to OP 
communication step just to cater for any, probably extremely infrequent, 
OP downtime?

=james.tindall

Andrew Arnott wrote:
> As I understand it, checkid_immediate is an optional step the RP can take,
> that was added for AJAX scenarios.  In my mind, this is a useless feature as
> checkid_setup also tends to be 'immediate' if checkid_immediate would have
> worked, and if checkid_immediate fails, then the very next thing the RP
> inevitably does is follow up with a checkid_setup request.
>
> As far as testing for OP-deadness, checkid_immediate is no good because it's
> a redirect of the browser just as checkid_setup is.  If the OP is dead, the
> user gets a dead-end error page regardless of what the openid.mode is set
> to.
>
> Now, assuming all OPs are alive, the idea of running through all endpoints
> with checkid_immediate first to see if any happen to authenticate, and only
> if that fails doing a checkid_setup on the first good OP in the list is an
> interesting idea.  It would serve the user as he/she would have a higher
> likelihood of not being prompted for credentials, which is cool.  On the
> negative side, it means that if the user has 3+ OPs listed, he gets
> redirected *4 times* before finally seeing his first OP's authentication
> page if none of them are willing to checkid_immediate.
>
> On Tue, Jul 15, 2008 at 9:16 AM, James Tindall <james at atomless.com> wrote:
>
>   
>> I've not double checked what the spec has to say about this Andrew but it
>> seems to me that if an association with the chosen OP exists and is alive
>> then the RP should simply attempt authorization using 'checkid_immediate'
>> and if that then fails the RP should try authentication using the next
>> endpoint in the list?
>>
>>
>>
>>
>> Andrew Arnott wrote:
>>
>>     
>>> Another thought: since a responsible RP only creates a single association
>>> with an OP and stores it until it expires, and these associations often
>>> last
>>> days, if the user has an endpoint in the XRDS doc that points to an OP
>>> that
>>> is currently down, but with whom the RP has an existing, unexpired
>>> association with, the RP shouldn't try to create an association with that
>>> OP.  Instead, it might say to itself "yes, I successfully have an
>>> association with this OP, so I'll redirect the user to it", but the OP
>>> happens to be down for 5 hours that day, effectively disabling the user's
>>> ability to log in, in spite of the multiple OPs listed in the XRDS doc.
>>>
>>> So... perhaps OpenID can have a "ping, are you alive?" message in its
>>> protocol.  But then we're no better than "dumb" RPs having to make
>>> multiple
>>> hits to the OP instead of just one.
>>>
>>> Another idea that keeps occuring to me is that the RP can use a frameset
>>> to
>>> keep an RP frame around and have the OP authentication happen in another
>>> frame (iframe or frameset would work).  That way, the RP frame could have
>>> something like "Is your Provider not responding?  Click here to try your
>>> next best choice..." or something to that effect.  The problem with this
>>> idea is that the URL on the browser will always be the RP, so the user
>>> will
>>> have one less way to confirm that this is indeed his genuine Provider and
>>> not one of those notorious OpenID RP phishing attacks.
>>>
>>> Anyone have any idea how to solve this dead OP problem?
>>>
>>> On Tue, Jul 15, 2008 at 1:52 AM, James Tindall <james at atomless.com>
>>> wrote:
>>>
>>>
>>>
>>>       
>>>> I've recently been writing an openid relying party library for the
>>>> Kohana php framework and I think you're right, the way I've tried to
>>>> handle multiple enpoints returned from the discovery phase is to first
>>>> sort them by their priority values and then also filter out any that do
>>>> not meet the (extension / security / openid-version) requirements set in
>>>> the current relying party configuration and then finally attempt to
>>>> establish an association with each endpoint in turn until a successfull
>>>> association response is returned.
>>>>
>>>> I feel that all this is best done invisibly rather than requesting the
>>>> user to jump through extra hoops. The user has after all -at some point-
>>>> set the priority of the endpoints listed in the xrds and I suspect that
>>>> most would not wish for further input during the process of
>>>> authentication. My assumption being that the point of multiple endpoints
>>>> is to hopefully cater for the requirements of as many different relying
>>>> parties as possible and to have alternative/backup endpoints incase of
>>>> errors or failures with any of the other endpoints in the list during
>>>> authentication?
>>>>
>>>> =james.tindall
>>>>
>>>> Andrew Arnott wrote:
>>>>
>>>>
>>>>         
>>>>> I'm curious how other libraries do (or plan to) handle multiple
>>>>> endpoints
>>>>>
>>>>>
>>>>>           
>>>> in
>>>>
>>>>
>>>>         
>>>>> a single XRDS document.  I see a few considerations, in order:
>>>>>
>>>>>   1. Enumerate the services in the XRDS-defined priority order
>>>>>   2. Skip the services that do not expose OpenID endpoints.
>>>>>   3. Skip the OpenID endpoints with Providers that do not quality
>>>>>   (whitelist/blacklist or advertised extension support
>>>>>   4. Take the first endpoint that is left after these filters.
>>>>>
>>>>> But what about the rest of the endpoints listed?  Here are some
>>>>> possibilities:
>>>>>
>>>>>   1. Just use the first endpoint and trust it works.
>>>>>   2. Try each one successively.  That is, the RP should attempt to
>>>>>   establish an association with each one until it succeeds with one, and
>>>>>
>>>>>
>>>>>           
>>>> then
>>>>
>>>>
>>>>         
>>>>>   redirect the user to that one for authentication.  Redirecting the
>>>>>
>>>>>
>>>>>           
>>>> user to
>>>>
>>>>
>>>>         
>>>>>   an unavailable Provider will result in a dead end failure page and the
>>>>>
>>>>>
>>>>>           
>>>> RP
>>>>
>>>>
>>>>         
>>>>>   will lose the opportunity at this point to try the next endpoint.
>>>>>   3. A variant on the last, except that in addition to skipping OPs that
>>>>>
>>>>>
>>>>>           
>>>> do
>>>>
>>>>
>>>>         
>>>>>   not respond to association requests, allow the user to "fail" or
>>>>>
>>>>>
>>>>>           
>>>> cancel the
>>>>
>>>>
>>>>         
>>>>>   authentication on the first provider and proceed to the second
>>>>>
>>>>>
>>>>>           
>>>> provider
>>>>
>>>>
>>>>         
>>>>>   listed for another authentication attempt.
>>>>>   4. Offer the user a list of his/her providers to choose from for
>>>>>   authentication.
>>>>>
>>>>> Have thoughts been written already on which of these are best and/or
>>>>>
>>>>>
>>>>>           
>>>> common
>>>>
>>>>
>>>>         
>>>>> in existing libraries?
>>>>>
>>>>>
>>>>> ------------------------------------------------------------------------
>>>>>
>>>>> _______________________________________________
>>>>> general mailing list
>>>>> general at openid.net
>>>>> http://openid.net/mailman/listinfo/general
>>>>>
>>>>>
>>>>>
>>>>>