Will the “real” federated provisioning please stand up?

Nishant has commented on my post about federated provisioning.  He has provided two different examples of federated provisioning.  One of these, the advanced provisioning example, involves a company who manages its employees’ access to a service provider service via provisioning.  In this case, Nishant agrees with me that provisioning of this sort is no different than provisioning the UNIX box down the hall.

 

But it is Nishant’s second example, the just-in-time provisioning example, which is a bit tougher.  In this case, the enterprise and its service provider have a federation in place.  Using SAML-based authentication, a new user attempts to access the service provider’s service.  The idea (hope?) is that the service provider recognizes the new user request, provisions the user, and authenticates the user in the same conversation. Nishant does add a degree of difficult in this scenario as he ties the federation service to a provisioning service.  Grabbing attributes from the SAML token, creating a SPML message, and handing that to a provisioning service is possible, but as a commentator points out this sort of interop isn’t spec’ed out so the heavy lifting is left to the service provider.  And even if the service provider doesn’t want to directly link its federation and provisioning services, it still needs to grab that assertion attributes and create the account in the backend system. 

 

It turns out, to my surprise, that there are people doing this.  Parties in a federation agree to which attributes are needed and send those in their authentication assertions.  A process at relying party uses those attributes to provisioning new accounts.  This is a fairly lightweight and effective approach, but there are some catches to be aware of.

 

The first catch, as Nishant points out, is if the service provider needs attributes above and beyond what are in the assertion, there’s not an easy way for the service provider to ask for them.  To deal with this, the service provider has to present a registration screen of some sort to the user.  Compared to the first scenario in which the federate account is already waiting for the user, the second scenario is herky-jerky and will annoy/confuse the end user.

 

The second catch is deprovisioning.  The provisioning process hinges on an authentication event.  Deprovisioning cannot be activated on de-authentication.  This does leave the problem of how to remove accounts when people have left a federation partner.  In the approaches we have seen, when a new account gets built it has an expiration date associated with it that gets updated on every login.  After some period of time without an authentication, the account is suspended or deleted.  Not a bad way to go.

 

JIT Provision may in fact be “real” federated provisioning, but not provisioning, as a dogmatic, dyed-in-the-wool provisioning guy would immediately recognize.  While I take my dogma for a walk, this quarter Lori and Bob are going to looking into some of the intersection point of identity management and SaaS and I think they’ll have more to say on this type of conversation in the coming months.

 

Down with federated provisioning

There’s been a bit of recent blogging activity about federated provisioning and SPML.  Having worked on both federated provisioning and SPML in a past life, it warms my heart to see this discussion.  Jackson, quoting the CIO of Education Testing Services, Daniel Wakeman, restates the observation that SaaS providers are providing when it comes to federated identity management.  This “major shortcoming” leaves service subscribers to fend for themselves in managing user lifecycle events like on-boarding and off-boarding.  Not acceptable.

That got me thinking – there really ought not to be a concept of federated provisioning.  Provisioning an application in the data center must be the same as provisioning an application in the cloud.  However, in the course of the conversation between JamesJackson, and Mark, it seemed SaaS applications and in-house applications were different from a provisioning perspective.

SaaS applications may be harder to provision and de-provision than non-SaaS application, but that doesn’t make them fundamentally different animals.  The point was made that SaaS apps lack a standards-based provisioning interface, an SPML interface.  The fact is the vast majority of applications, SaaS or not, lack a standards-based provisioning interface and this makes dealing with them very much the same.

Now there are two reasons that we don’t hear the same short of clamor about provisioning non-SaaS applications as we do with SaaS applications:

  • We’ve dealt with it so long that pain isn’t as acute
  • Provisioning vendors built an array of connectors, shifting the pain up a level, allowing companies to focus on the provisioning technology and not each and every application they want to provision

Provisioning vendors spent lots of time and money to build connectivity to traditional applications.  Lots.  And in doing so provided a bit of absolution for application vendors from their failing to provide a standards-based provisioning interface.  Having gone through all that pain and suffering, vendors are not eager to go through it again with SaaS applications, coding connectors to each one’s different web service.  Customers aren’t too keen on the idea either.

In providing SPML interfaces to their applications, SaaS vendors would do everyone a service.  Provisioning vendors could use their SPML connectors and not have to build to each SaaS application.  Customers wouldn’t have to write custom code to different service interfaces.

You don’t want that fired sales guy walking away with your customer list no more than you want him walking out the door with your pricing information.  To that end, there should be no reason why deprovsioning from an application like Salesforce.com is any harder than deprovisioning from LDAP.  Federated provisioning should not exist; there is only provisioning.

(Cross-posted from Burton Group’s Identityblog)