Add back the SharedTokenCacheCredential to handle token which is cached by the InteractiveBrowserCredential

[ArthurKamalov]

The original problem: the InteractiveBrowserCredential required login by browser on each new run.

Observations:
It seems that the InteractiveBrowserCredential is able to save its token to persistent cache storage, but it is not able to use that cache later. ( It actually has its own method _acquire_token_silent, but, as far as I can see, it does not check for token in cache, but already expects an instance of the AuthenticationRecord class, which, I assume, we have to handle and store manually)
That's why I also made an assumtion that we should use SharedTokenCacheCredential for that. I know that the SharedTokenCacheCredential was previously removed by @vxfield and there might be a valid reason for that, but I haven't found any better workaround, so I would glad to hear whether we can use it or not.

[ArthurKamalov]

On Linux, when passing self.tenant_id to the SharedTokenCacheCredential, it fails due to a mismatch of this tenant_id with tenant_id from the cache. Still haven't figured it out why. It seems that it is able to work without it, but I am nearly sure that may cause additional problems too.

[vxfield]

even if the tenant_id is not passed, we automatically fetch it anyway by calling self._discover_tenant_id_()

[ArthurKamalov]

But that's for the constructor of the _DefaultAzureCredential, right? I meant passing the tenant_id to the SharedTokenCacheCredential itself. Is there a chance that the SharedTokenCacheCredential will pick a wrong data without specifying the tenant_id

[vxfield]

Also, since the InteractiveBrowserCredential already has support for caching, I'm confused why we need to add the SharedTokenCacheCredential in the list of credentials.

[vxfield]

Ok, I think I get it.
If we enable the caching in the InteractiveBrowserCredential it will fail even if the browser authentication would succeed but accessing the cache fails.

[vxfield]

Please see my latest commit.
Using our own cache will mitigate issues with tokens cached by other apps.
As Kirill mentioned, if there was a change of authority oder tenant_id, the SharedTokenCacheCredential will be skipped and the user can authenticate with the InteractiveBrowserCredential (which will store a new token for these new parameters).

[vxfield]

Observations:
It seems that the InteractiveBrowserCredential is able to save its token to persistent cache storage, but it is not able to use that cache later. ( It actually has its own method _acquire_token_silent, but, as far as I can see, it does not check for token in cache, but already expects an instance of the AuthenticationRecord class, which, I assume, we have to handle and store manually)
That's why I also made an assumtion that we should use SharedTokenCacheCredential for that. I know that the SharedTokenCacheCredential was previously removed by @vxfield and there might be a valid reason for that, but I haven't found any better workaround, so I would glad to hear whether we can use it or not.

The _auth_record is automatically set upon successful Browser Authentication source.
We don't need to worry about it if we also use the SharedTokenCacheCredential as you said to use the cache on a different process or instantiation of our SDK.

[kikomiss]

I think we might also need to pass tenant_id=self.tenant_id to here. It will allow the filtering part to pick-up accounts only the discovered tenant

[vxfield]

done

[ArthurKamalov]

As I mentioned in this comment #603 (comment), on Linux I have a mismatch of a current tenant id and id from the cache, which leads to exception in the SharedTokenCacheCredential. Still trying to find the cause, still not sure if that's only on my machine or not.

[vxfield]

@ArthurKamalov, @kikomiss: I'll hand this back to you now.

I think the next steps would be to add test cases for the _DefaultAzureCredential.get_token() method with different authentication scenarios.

See how we instantiate the _DefaultAzureCredential here.

These new test cases:

• Can be added in the test_authentication.py file.
• Can mock the get_token() method of each credential that we support and check if _DefaultAzureCredential.get_token() is returning that token.
• Can mock the _get_persistence function to raise an Exception and check if a Warning is emitted during the _get_cache_options() method, and that it still returns the mocked InteractiveBrowserCredential token.
• etc...

[ArthurKamalov]

Sorry if I am missing something here, but should this be an error log level?

and what do we really guard here in that try/catch clause? cache._persistence.get_location() ?

[vxfield]

Good catch. I was using error log level for local testing. I updated it to info.

I guarded the cache._persistence.get_location() because:

1. We are calling it just for tracing / info logging purposes
2. _persistence is a private attribute that can change in the future
3. get_location() could raise an expected exception in a particular situation that we are not aware of
4. The cache implementation is different for each OS and would be hard for us to guarantee it always work
5. Finally, I thought it's not worth the risk of this info logging to break our auth :-)

But feel free to modify the code as you see better.
Like we could try to get the location, but even if fails, it we could log the rest of the info ("Using Azure.Identity Token Cache.").

[vxfield]

I made a small improvement based on my own comment, but feel free to adjust it.

[ArthurKamalov]

Yeah, that makes a lot of sense now, thanks :)

[vxfield]

@kikomiss, hopefully this can provide additional guidance for users facing issues with libsecret, but it's a best effort try, and I like your idea of having some documentation in Microsoft Learn docs.

[vxfield]

If we want to check if libsecret is working (instead of checking a ValueError message that can change in the future) we could call msal_extensions.trial_run() if sys.platform.startswith("linux").

[vxfield]

Again, I implemented my own comment :-)