Atomic Workstation (Was: Re: Call for agenda for Workstation WG meeting 2015-Sept-02)

[Owen Taylor]

On Tue, 2015-09-08 at 10:55 -0600, Chris Murphy wrote:
> On Tue, Sep 8, 2015 at 10:00 AM, Owen Taylor <otaylor at redhat.com> wrote:
> 
> > *However* - I think it's important to realize that the interesting and
> > hard part of the project here is *not* using rpm-ostree for the base
> > operating system. Putting Fedora Workstation images into an ostree
> > could be done today; getting installation and updates to work is a few
> > weeks to a few months of work but something we could very plausibly
> > have done for Fedora 24. But what would we have then? Something that
> > almost nobody would want to use ... certainly not developers. We'd have
> > taken away the ability to freely mutate the operating system, and
> > provided any sort of replacement.
> 
> I have thought abstractly about and interim approach by leveraging the
> part of ostree that understands the "trees" state of an OS, and
> therefore manage the bootloader configuration, without the rigid
> immutability aspect you refer to. That's a derivative of ostree +
> lvmthinp and/or Btrfs snapshots. That gets us rollback with
> conventional mutability but can also help limit slew.
> 
> That might be too much work for its implied limited lifespan until the
> app dev and deploy angle for Atomic Host is sorted out and matured. It
> also might overstate the importance of rollbacks. Afterall, we've
> never had that and the world hasn't ended.

Maintaining a local rpm-ostree repository and using that to keep a
history of the states of your system is certainly possible, and
possibly likely useful in some circumstances.

But on one hand, you still can't, for example, update one package
without rebooting. And on the other hand, it doesn't provide the
biggest advantage that we can offer to users: the assurance that we've
actually tested not just individual packages that we're installing on
the system, but the actual same operating system that they are running.

- Owen