[Container-tools] Notes from Container Tool calls July 1st & 2nd

[Daniel Veillard]

           Some notes from the Container Tool Cabal call
		 and Container Tool standup
		  July 1st & 2nd 2015

A number of people were on vacations, here is a log of the
notes I tried to gather while running the call, likely very
incomplete, feel free to chime in and correct :-)

Dusty Mabe joind the call for the first time, he will be with us
from August but join in to smooth his transition from consulting role.

Atomic community booth at the Red Hat Summit last week:
   - it was very good to have a booth close to the entrance of the
     community room ! Too bad the community room itself was a bit
     far away. The ARM gizmos running CentOs were like visitor's
     magnets, highly recommended :-)
   - most of the focus was on containers and associated generic
     questions, maybe a hundred visitors, most of the questions
     were rather basic about containers but a few came to discuss
     Nulecule.
   - Seems like the Foreman people might be interested in Nulecule for
     deployment
     the relation w.r.t. existing OpenShift tooling wasn't always well
     understood
   - It takes an awful long time to explain what we do, how can we
     reduce the 10mn explanation down to a 2mn one ?
   - Seems people are still undecided, nulecule would get more adoption
     if it was clearly seen as part of the whole ecosystem
   - storage issues, for those who tried, the default of using the
     device mapper in loopback mode kills performances, we need to
     make clear how to set up perfomant storage

Documentation and terminology:
   - there is confusion around the Atomic term, is a really nice term,
     but used by a set of component who can be used separately. There
     isn't really a big picture that we can show to people and they
     would instinctively see how everything is framed together.
     + Atomic Host
     + Atomic Platform/Enterprise
     + Atomic app
   - As Mark put it:
     "I want a scale level diagram.  Docker/rocket/lxc down on a host,
     next level the host and the dockerd itself, then kube doing host
     management, and forming clusters of hosts running groups of single
     containers, then Atomic Plaftorm and OSE helping devs and admins
     do that, automate, visualize, then finally Nulecule defining apps
     as sets of cooperating containers"

Demos:
   - we need to improve the Pulp demo, add encryption and authentication
     support to be actually usable in the field
   - need to be careful, Kubernete port 8080 access likely to be removed
     in coming release so need to plan if someone uses it
   - the issue with kubernete not starting on recent build found, the
     service decription need to indicate that network start from systemd
     has finished before starting it. KB get fix from Lala should be able
     to update builds.

Improving nulecule flexibility:
   - right now parametrization relies on environment variable substitution,
     this is problematic in multiple ways and rather limited
   - before shipping a supported version we should implement a more
     powerful mechanism based on what is basically XPath for JSon
     https://github.com/projectatomic/nulecule/issues/70
     DV being an Xpath implementor but for XML will have a look, but
     no hard promise

CentOS 6 story:
   - a (very) large majority of the people deploying CentOS are on CentOS 6,
     so people ask for it.
   - in upstream RHEL the whole effort is supported only on RHEL-7
   - Docker ships a CentOS 6 version but it fails by default since one need
     an updated kernel, one need to update to the kernel provided by
     the Virt SIG to get things working
   - We can't change the CentOS 6 releases done by Docker, at least we
     should push documentation

Daniel

-- 
Daniel Veillard      | Open Source and Standards, Red Hat
veillard at redhat.com  | libxml Gnome XML XSLT toolkit  http://xmlsoft.org/
http://veillard.com/ | virtualization library  http://libvirt.org/