Installation Roles vs. Post-installation Role Assignment (or pets vs. livestock)

[Simo Sorce]

On Sat, 2013-11-16 at 02:17 +0100, Miloslav Trmač wrote:
> > Or they can just install a single role into a container or a run it
> in a VM
> > on the " *one* beefy machine"
> 
> And... what OS does the "one beefy machine" run?  Is it a pet or
> livestock?

[strawman: Let me replace livestock with 'bee', as proposed by my
colleague Adam, bees are really throwaway from the Hive :) ]

My point is even simpler, the bees idea can work for some software,
usually purpose-built by the mentioned DEvOps to be throw-away friendly,
where all the state is kept on a separate redundant database or
filesystem and so on.

We should work to make Fedora as friendly as possible to those
workloads.

But most of the software we have now is not throw-away that way, so
concentrating on a system that only cares for bees would miss the point.

Whether a Server is a pet or a bee is a choice of who installs it it
should work well in both roles, as well as we possibly can.

Of course you can't expect to install 15 different roles on the same
machine and expect nothing conflicts and all performs great, but that is
not the point either.

We build roles as collections of software and reasonable defaults in
configuration so that a collection of programs work together to perform
a well defined task. In some cases multiple roles will be able to
coexist (database role + web server role for example).

In some other cases roles will conflict. In that case it will be nice to
be able to instantiate roles in different containers if possible, for
example, but in order for that to happen containers will need to improve
quite a lot, I do not think we can depend on that as a given yet.

Simo.

-- 
Simo Sorce * Red Hat, Inc * New York