Excerpts from Amit Saha's message of 2014-08-20 14:04 +10:00:
1. If we do not allow selecting the base image, we are effectively
*not* obeying
the distro selection in the job unless of course both are the same.
Right, so I think there are just two parameters which were hardcoded in
your experiment but which need to be adjustable if we implement the
feature for real: Docker base image to run the harness inside of (e.g.
fedora/20), and Docker registry URL to pull from (expectation being that
the user will supply a lab-local one to avoid pulling huge images over
the WAN).
To begin with we could just have ksmeta variables for those both. It's
not great (no way to automatically pick a local registry automatically
for each lab, no way to make Beaker pick the right distro using filter
criteria like <distroRequires/>, etc) but it would be okay as a first
cut.
2. If we start the container with /run volume mounted, running other
containers on
the host will be possible and so will be restarting the host. We *could* make this
always the case or have an option to enable it.
For things like this I think we should just pick some sane defaults and
leave them like that to start with. Privileged container, as little
isolation as possible, and any useful filesystems bind-mounted (/run and
/etc at least).
--
Dan Callaghan <dcallagh(a)redhat.com>
Software Engineer, Hosted & Shared Services
Red Hat, Inc.