[Libosinfo] unattended installation support

[Fabiano Fidêncio]

On Tue, 2019-06-04 at 10:38 -0400, Cole Robinson wrote:
> On 6/4/19 3:30 AM, Fabiano Fidêncio wrote:
> > People,
> > 
> > libosinfo/osinfo-db claims to support unattended installations of
> > various OSes and all the possible versions of each OS. However,
> > this
> > is not tested (apart from people using it on Boxes).
> > 
> > There are a few things that come to my mind:
> > - Having tests for unattended installations:
> >   - First step towards this has been done with the unattended
> > installation support being added to virt-installl. A lot more work
> > will need to be done in order to actually have it tested in a sane
> > way
> > ...
> 
> Yeah. There's definitely opportunity for some unit style tests in
> virt-install, but this stuff will need a functional test suite in
> some
> form, and some of it non-public for the windows stuff at least.

Yep. And the way I see a functional test suite being implemented is
usig virt-install to do so.

> 
> > - Drop support for EOL systems:
> >   - We claim to support Fedora >= 3. Was this ever tested?
> >   - We support Windows >= 2k. This was tested back in 2012 ... but
> > is
> > this something we really should keep?
> > > My suggestion (and I'd like to get the others maintainers bless)
> > > is
> > to, right now, only keep support for:
> > - Linuxes: The supported versions of the distros (so, for instance,
> > for Fedora ... just support Fedora29 and Fedora30);
> > - Windows: Windows 7+
> > 
> Dropping win2003 and earlier will simplify app implementations
> because
> they can (probably) avoid the floppy requirement, like we discussed
> on
> IRC. So I think that helps quite a bit. There's still definitely
> people
> using winxp (despite being out of support microsoft just pushed a
> patch
> update last week) but I don't think we need to support unattended
> installs there anymore.
> 
> My opinion is mixed on killing off unattended support for EOL
> distros.
> Doing a one time purge now is one thing, but going forward is the
> more
> interesting bit. It would mean that working virt-install command line
> invocations may break in the future which I never like. But I don't
> want
> osinfo to be restricted by maintaining unattended compat with a ton
> of
> old distros...
> 
> Maybe we can have an informal tier system based on what we test. So
> when
> making unattended changes to git, we only functional test the non-EOL
> distros. When a distro goes EOL we can fork off the install script to
> fossilize it, so changes we make for Fedora 30 don't alter the config
> for EOL Fedora 28.
> 
> Presumably if we do a one time test of say Fedora14 installing
> correctly, then we fork the unattended install script, we can have
> reasonable confidence that it will continue to work into the future
> unless things change on the Fedora side. But we only do full install
> testing with latest supported Fedora. But if for example fedora14 is
> already broken then I'd say just drop it

Yep, I'll end up trying all unattended installations on all old entries
we have and figure out what actually works, what should be improved and
what are the improvements need.

Daniel, although not in this thread, has a quite clear opinion on *not*
dropping any support on osinfo-db side. So, although I'd prefer to say
to apps "stick to this version of osinfo-db" or "add this bits locally
if you need", I totally understand the reasoning behind Daniel's wish.

> 
> - Cole
> 
> > By taking this path we'll have an easier time to ensure that our
> > scripts are working and also reduce the maintainability of those.
> > 
> > Shall I go for this change?
> > 
> > Best Regards,
> > 
> 
> - Cole


Best Regards,
-- 
Fabiano Fidêncio