Code review comment for lp:~jimbaker/juju-jitsu/unit-test

On Wed, Nov 14, 2012 at 1:10 PM, Benji York <email address hidden>wrote:

> Thanks for this subcommand, Jim. Nicola and I have been using it to write
> some tests (see a first cut at lp:~juju-gui/charms/precise/juju-gui/trunk).
> A few remarks follow.
>
> The charm directory name has to be the same as the charm name itself,
> which makes testing different branches difficult. Is revisiting that
> requirement possible?
>

This is a core requirement to guard against errant charm authorship. Its
not a jitsu check per se, but a juju core one. Its unlikely to be revisited
imo, but #juju-dev would be the appropriate place to confirm that.

>
> It seems to us that the exit code from each one of the *.test files should
> be propagated and reported by the test subcommand (e.g., if any test
> returns a non-zero then Jitsu should return non-zero).
>
> Bootstrapping from within the command either causes errors or times out:
> one needs to bootstrap in advance and use the --no-bootstrap option.
>
> When accessing machines for log retrieval, rsync asks for a password for
> ubuntu@localhost. We can not figure out what that password should be.
>

try setting up the key of your environment file as an authorized key for a
user ubuntu.

i assume based on the address that this is being tested with the local
provider, which is a bit of an oddball compared to other providers as the
default user isn't 'ubuntu' but whatever the user account is. ideally the
test runner could distinguish this based on provider type and use the local
provider user account. that would still need manual addition of the ssh key
to authorized keys as the local machine is not cloud-init initialized.
alternatively in the case of the local provider the testrunner eschews
machine log collection in favor of just unit log collection which should
still work as per other providers.

 hth,
Kapil