The syntax used in this execution was flawed.

This is a QoL change to help us fine tune the artifacts in CI.

We were using only the base name to create the image file, which could
cause conflicts if we ever try to setup both releases.

The 'await' command on virsh is not available on Noble, which is our
runner in github, so we need to use another way to check for the domain
status.

Update the commands to also run `sudo apt-get update` and add some
dependencies that were missing from the lists

This workflow downloads an Ubuntu image, sets up some of the
aplications and configurations that we will need in the VM in order to
be able to run the E2E tests and then caches it for later use to save us
some time when running the tests.

Co-authored-by: Adrian Dombeck <[email protected]>

fix output artifact naming

ci/e2e-tests: Use env vars instead of config file

The config file is useful when running the tests locally, but it only
sets environment variables, so in the CI it's simpler to just set the
environment variables directly.

ci/e2e-tests: Fix indentation

This caused the provisioning to fail

We were padding the generated code with trailing 0s, but we should have
been padding with leading 0s. This commit fixes that.

The code is generated according to the current time and is valid for 30
seconds. This means that if we generate the code just before the time
window changes, it might be invalid by the time we use it.
To avoid this, we make sure the time is safely within a new window before
generating the code.

We see permission denied errors when trying to start the VM both in CI
and locally because AppArmor denies access to the qcow2 image
(curiously, starting the VM by restoring a running snapshot works).

We should investigate the issue and report it upstream. In the meantime,
let's work around it by disabling AppArmor confinement for the e2e-test
VM. That should be fine since we're not doing anything in the VM which
would make them more vulnerable than the host.

We recently changed the error message that is shown when the broker
fails to reach the provider, so we need to update this check on the
tests as well.

Sometimes machinectl login hangs and it causes the tests to fail. Let's
try reopening it a couple of times (in case of failure) to see if we are
able to get more consistent behavior

systemd only supports sending the journal via VSOCK since v256. In the
GitHub CI the newest Ubuntu release we can use on the runners (both
GitHub-hosted or self-hosted) is 24.04, which has systemd 255.

Running the tests is quite resource intensive, so let's switch to using
our self-hosted runners to see if we can get some improvements in
performance and stability.

It doesn't only "Build and Cache Ubuntu VM", it runs the e2e-tests and
does everything required to do so.

On GitHub Runners, the VM can't connect to the host via TCP. This commit
instead makes the host connect to the VM to retrieve the journal.

On the self-hosted runners, the tests fail because the SSH connection to
the VM fails.

Also, downloading the VM image from the artifacts takes 12 minutes
instead of 20 seconds.

This reverts commit 4201248.

The tests are also useful when targeting other branches than main.

There are not many files in the repo which don't affect the e2e-tests
(once we run them with authd and broker artifacts built from the branch)
but it's still worthwhile to add include/exclude paths because the tests
are very resource intensive.

To avoid duplication between the msentraid and google e2e-tests.

So that we automatically re-provision the VM image if anything changed
in the relevant files.

It's too much output

It's not only testing authd-oidc-brokers but also authd.

We ignored errors during cloud-init and happily continued provisioning a
broken VM.

Should save some time and bandwidth.

The file is already executable