[Pulp-list] Fwd: CDS: gofer authentication
jason.dobies at redhat.com
Fri Mar 18 16:40:27 UTC 2011
-----BEGIN PGP SIGNED MESSAGE-----
One more thing:
secret = self._cds_stub(cds).initialize()
except RequestTimeout, e:
raise CdsTimeoutException(e), None, sys.exc_info()
except DispatchError, e:
raise CdsCommunicationsException(e), None, sys.exc_info()
except NotAuthorized, e:
raise CdsAuthException(e), None, sys.exc_info()
except Exception, e:
raise CdsMethodException(e), None, sys.exc_info()
NotAuthorized is a subclass of DispatchError, so that code block will
never get executed. Need to reverse the order on those.
On 03/18/2011 12:38 PM, Jay Dobies wrote:
>> I pushed the pulp -> external CDS authentication feature.
>> Commit: cf449c22299e01e872dff9b92a3cff65ac39cc36
>> Mind doing a quick review? I'll be making a pass though the unit tests
> Looks pretty solid. Most of my comments focus around things you'll run
> into in one way or another when you do the unit tests, though you may
> have approaches in mind and I'm just not thinking of.
> - init_cds(): Method now returns the secret, don't forget to update the
> docstring to mention that and say if it's a string or something more
> - There aren't any tests for this (sorry about that, my bad) so you'll
> have to add the test file itself.
> - How are you going to override the location of the secret file for unit
> tests? The path can be specified but the object is constructed inside of
> getsecret(). The default is to read from config, but that's loaded by
> gofer (there may be some way of overriding config values loaded by
> gofer, which is one reason I'm asking since I don't know).
> - We'll need tests that run both with the secret file present (scenario
> of normal usage) and not present (scenario of an initialize). When
> cleaning up the test runs by deleting the secret file directory, won't
> we run into an issue since Secret is a singleton, is already loaded, so
> __mkdir won't get called again?
> - The caching doesn't come into play if gofer is restarted. The secret
> is cached on write, but not on read. So if the CDS is bounced after it's
> initialized, the caching isn't used.
> - In my experience, singletons have always been a headache in unit tests
> since modules aren't unloaded between runs, so you never know exactly
> which test is the one causing the instantiation. Is there a benefit to
> this approach over simply having module methods read_secret,
> write_secret, and delete_secret and caching in the module itself?
Pulp-list mailing list
Pulp-list at redhat.com
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.14 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org/
-----END PGP SIGNATURE-----
More information about the Pulp-list