canonical-ci-engineering team mailing list archive

[Evan Dandrea <evan.dandrea@xxxxxxxxxxxxx>]

On Wed, 28 Oct 2015 at 04:58 Olivier Tilloy <olivier.tilloy@xxxxxxxxxxxxx>
wrote:

> Hi,
>
> There used to be a thing called 'otto' that ran autopilot tests on
> desktop (in VMs) as part of the CI process on merge requests.
> IIRC this didn’t work very well and had known shortcomings, so it was
> retired.
>
> Is there a plan to add back something similar that would run autopilot
> tests on desktop?
> With our focus on convergence, the current CI setup leaves out a
> number of features untested, not because the tests don’t exist, but
> because they are not being run automatically (we do run them manually
> for the time being, but this doesn’t scale well).
>

Hi Olivier,

Your description of Otto is accurate. It broke hard whenever the kernel
moved and proved to be untenable.

The CI team has been in wind-down mode since the UES re-organisation back
in May, with a target of end of 2015. As part of this transition, they've
been making it possible for engineering teams to get a production Jenkins
server with minimal effort [1]. Meanwhile, the Certification team has taken
over responsibility for providing testbed hardware as a service to UES
teams.

The end result is that you can file an RT and IS will deploy a Jenkins
server (and slaves) that your team has complete job control over. You can
ask Certification for phones, bare metal, or whatever specialised hardware
your need to run these jobs on, and they will provide a Jenkins slave that
can be plugged into your Jenkins server for said purpose.

You can find comprehensive documentation for "Jenkins as a service" here,
complete with example jobs for common tasks:
https://wiki.canonical.com/InformationInfrastructure/Jenkaas/UserDocs

Do note that while IS will ensure that the Jenkins server process keeps
running smoothly and Certification will resurrect testbed hardware that
gets wedged, the engineering teams are themselves responsible for writing
and debugging their Jenkins jobs, as well as test results.

If you have any questions, feel free to ping me on IRC or stick something
in my calendar.

Thanks,
Evan

1: Most of our requests were of the form "please may I have a Jenkins job
to ..." It was determined we could best help the engineering teams before
parting ways by making it push-button to obtain infrastructure to run jobs,
process MPs, and talk to testbed hardware.