Re: [atomic-devel] Systemd, containers, and pid=host

[Nick Coghlan <ncoghlan gmail com>]

On 12 March 2015 at 06:14, Clayton Coleman <ccoleman redhat com> wrote:
> Oops, I meant "outside of containers so that people can easily move simple apps into containers"

This is something I'd really appreciate. Currently, my container
adoption plans including reverting to traditional virt at the
independent QA stage of a deployment pipeline, as containerising local
developer workflows is a lot easier than containerising the entire
production environment.

Right now, Vagrant is actually a better fit on the development side,
since that gives full VMs where systemd service files will work as
expected.

So even though I know it's doing containers "wrong", I'd really like
to be able to swap out vagrant-libvirt VMs for lighter weight Docker
containers, without needing to make the apps themselves container
aware (at least not yet)

Once that "big container" model was working, *then* we could look at
breaking things up to achieve the "one service, one container" goal,
with something like Kubernetes for grouping.

But jumping straight from "multiple services on a VM" to "multiple
containers in a pod" sounds like a recipe for some nightmare debugging
sessions compared to the more incremental: multiple services in a VM
-> multiple services in a container -> start moving services
incrementally out to other containers in the same pod -> look closely
at the services to see which ones really need to be collocated and
which could be moved out to different pods entirely.

Regards,
Nick.

-- 
Nick Coghlan   |   ncoghlan gmail com   |   Brisbane, Australia