Impersonation for non-persistent authenticators

[segy]

Hi

we were planning to implement impersonation for a stateless authenticator and it looked logical to me, to implement ImpersonationInterface, so we can use auth component for that. But this interface extends PersistenceInterface. I don't see any logic in this inheritance. Could you please explain, if this is needed? If not, I can create PR for the necessary changes.

Thank you

[ADmad]

How do you expect impersonation for stateless authentication to work?

[segy]

we have this scenario (simplified) for JwtAuthenticator:

• we have user encoded in the token
• when we start the impersonation, the service will return a modified token containing also impersonation information
• when he stops, it returns the standard token

[segy]

and the thing is, i couldn't find any reason for ImpersonationInterface to extend the PersistenceInterface. even the SessionAuthenticator (that is the only class implementing this) has a code implements PersistenceInterface, ImpersonationInterface, so the only change here would be to remove extends PersistenceInterface.

[markstory]

when we start the impersonation, the service will return a modified token containing also impersonation information
when he stops, it returns the standard token

Where will you store the standard token if authentication is stateless?

[segy]

nowhere. if we carry information about the impersonator, we can simply issue a new one when he decides to stop impersonating.

[markstory]

nowhere. if we carry information about the impersonator, we can simply issue a new one when he decides to stop impersonating.

So how would stopImpersonating and isImpersonating work? If you just want to manage the impersonation credentials yourself you should be able to do that already.

[segy]

Of course I am able to do that. But sorry... I don't understand, what we are talking about :) If you insist, that the ImpersonationInterface extending PersistenceInterface is necessary and you don't want to change it, just close this.

[segy]

but to answer your question: the isImpersonating obtains the request (and the token, which contains this information, is in the header). the impersonate and also stopImpersonating allows us to modify the response, so this logic can be handled in the authenticator.

[ADmad]

I don't understand, what we are talking about :)

Whenever someone opens an RFC we first try to understand the use case before we take a decision on whether to go ahead with the changes, hence the questions.

[segy]

i added a proposed change, so you can consider.

[markstory]

But sorry... I don't understand, what we are talking about :) If you insist, that the ImpersonationInterface extending PersistenceInterface is necessary and you don't want to change it, just close this.

Part of the ImpersonationInterface is that it remembers what your identity was before the impersonation began, and can restore that with stopImpersonating and check whether or not a user is being impersonated with isImpersonating.

I'm interested in how you see these method working with stateless authentication methods? For example basic auth or token based auth? Is that behavior you are expecting from the framework or are application authenticators that implement the ImpersonationInterface going to need to figure that out?

[segy]

OK. But do you think, that the fact, that it will be impossible for some authenticators to implement ImpersonationInterface, is a good reason for it to extend PersistenceInterface? I think that this is a responsibility of the authenticator that implements it to handle all the methods that are required by the interface. What is the benefit of that inheritance in this case?

[markstory]

I think that this is a responsibility of the authenticator that implements it to handle all the methods that are required by the interface. What is the benefit of that inheritance in this case?

I agree that each authenticator will need to manage the state themselves. Having the inheritance formalizes that there needs to be some persistent state managed somewhere, and gives application developers the necessary hooks to have state clearing integrated well. For example AuthenticationService::clearIdentity both resets impersonation and removes the identity.

Impersonation requires some state to be persisted somewhere and having an authenticator pretend it is stateless but also have persistent state feels awkward.

[segy]

so even this is possible to achieve with JwtAuthenticator, you think it's not worth merging?

[segy]

please let me know. i wanted to have this process unified using component and if this is not merged, i'd need to think of other solution. thanks

[markstory]

 so even this is possible to achieve with JwtAuthenticator, you think it's not worth merging?

This isn't possible with the JwtAuthenticator today. Or are you saying it could be done with JWT authenticator.

[segy]

Yes. I wanted to say it could be done.

[segy]

so please can we somehow close this? either as accepted or rejected? thank you

[markstory]

so please can we somehow close this? either as accepted or rejected? thank you

My issue with this change is that removing the PersistentInterface isn't very logical as if we do, the authenticator needs to have state stored somewhere for the component/service logic to work. If that state is required, what benefits are there to storing it somewhere else?

so even this is possible to achieve with JwtAuthenticator, you think it's not worth merging?

If you're proposing we have a stateful JwtAuthenticator that enables impersonation, then that is something that is more interesting. As a framework I think we should be helping enable common patterns, and having JWT authentication alongside an impersonation mechanic fits nicely with those goals.

Edit: To clarify, the 'state' could be held within the JWT token and not elsewhere.

[ADmad]

To clarify, the 'state' could be held within the JWT token and not elsewhere.

That's what he suggested in an earlier comment #648 (comment).

When impersonification is started they generate a new token which contains additional info regarding the user being impersonated and on stopping revert back to a "normal" token.

How the token switching is done can be left to the users and we don't need to do anything extra in the plugin, just update the ImpersonationInterface as suggested.

[markstory]

When impersonification is started they generate a new token which contains additional info regarding the user being impersonated and on stopping revert back to a "normal" token.

Ok. Good to see that we're thinking about the token switching in a similar way.

How the token switching is done can be left to the users and we don't need to do anything extra in the plugin, just update the ImpersonationInterface as suggested.

Shouldn't we at least provider an implementation of JwtToken with Impersonation? Having looked at the code further and talked through how the impersonation would work with a stateless authenticator, I am sold on removing the inheritance with PersistenceInterface.

[ADmad]

Shouldn't we at least provider an implementation of JwtToken with Impersonation?

I am not opposed to it if someone wants to provide an implementation :)

[dereuromark]

So is this still to be kept open? The related PR got merged.