ifcfg.j2 has a pretty awkward assumption that the interface
connected to the infra network will be eth0 (or enc900) - it
only includes the GATEWAY, DOMAIN and DNS1/DNS2 lines if the
interface is one of those two. It seems we were trying quite
hard to make eth0 always be "the interface", but now that's
been broken in a few systems. enc900 was added as apparently
that's what the main interface is called on some s390 boxes;
on openqa-ppc64le-01 the if that's connected is eth2 (eth0 is
present, but not connected), and on the new qa01 and qa02, it's
em3 (according to smooge, we have to use 'predictable' interface
names on those boxes as the old names really *do* get assigned
to different interfaces on each boot).
So since we now have several different cases where the 'eth0'
assumption doesn't hold, let's build a slightly better system
for handling it. This replaces ifcfg.j2's hard-coded list with
a variable, and sets the default value of the variable to the
two names ifcfg.j2 handled before: [ 'eth0', 'enc900' ]. This
allows the systems where the main interface is *not* one of
these to set the variable accordingly, and hopefully that'll
give them correct ifcfg files.
This *should* solve the problem of openqa-ppc64le-01.qa and qa01
and qa02 constantly dropping out of network connectivity any
time they got rebooted or the network plays got run.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
Hopefully this bit in tap-setup.yml can now go away, as this
approach of using ansible_ifcfg_whitelist and _disabled does the
same thing in a cleaner way.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
...except ppc64le, which seems to have dropped dead. Once that
one is done, we can drop the ugly bits from the tap-setup play,
I hope.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
qa01 seems to be stuck in network hell and we can't fix it
apparently. So make qa09 tap worker again for now.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
We can't have two tap worker hosts per instance as then openQA
assumes it can share related jobs between them, which we're not
set up for. Also the new boxes seem to be having some issues
with tap jobs. So let's make qa14 the only tap worker in prod
for now, and keep qa01 the tap worker for staging so we can
work out the problems.
Signed-off-by: Adam Williamson <awilliam@redhat.com>
