On Fedora 41 /etc/systemd/journald.conf isn't shipped by default anymore.
Instead /usr/lib/systemd/journalc.conf is the default
On RHEL and older Fedora's we can just not set this and it will
find the default one.
Signed-off-by: Kevin Fenzi <kevin@scrye.com>
In Fedora libvirt switched to a collection of socket activated services
instead of a large single libvirtd service. Restarting that service
caused the socket activated ones to die and then libvirtd exited on
idle, leaving nothing handling virt connections.
So, lets stop trying to restart that and no need to notify other things
because they should idle out/restart on socket activation.
Signed-off-by: Kevin Fenzi <kevin@scrye.com>
In f41+ libvirt defaults to using nftables if both it and iptables
are installed, but it doesn't seem to work with imagefactory/oz
virt instances and our iptables setup.
So, lets revert back to iptables for now.
We can switch back if we can fix the incompatiblity, switch builders
to nftables, or stop using oz/IF
Signed-off-by: Kevin Fenzi <kevin@scrye.com>
dnf5 requires neither python3-rpm nor python3-dnf, so we cannot
assume they are installed (as they always were before). ansible's
package_facts needs python3-rpm (even in the latest upstream
version), so we at least need that. The ansible in rhel9 (which
is what's on batcave01 currently) does not support dnf5 at all,
so we need python3-dnf installed so it can use its dnf3/4
support. python3-dnf relies on python3-rpm, so we can just
check for and install python3-dnf. Once batcave01 is updated to
an ansible that supports dnf5 we won't need python3-dnf, but we
*will* still need python3-rpm.
Stuffing this in yumrepos.yml as it runs on everything, early,
and is packaging-related. We could make it a separate task but
then we'd need to make every playbook include it...
Signed-off-by: Adam Williamson <awilliam@redhat.com>
Prior this commit, the script flushed the content of the osbuildapi
set before adding the newly resolved IP addresses into it. Even worse,
the DNS resolve for sso.redhat.com happened after the flush, so there
was no IP address for sso.redhat.com for quite some time (resolving
sso.redhat.com without caching takes 30 ms on my machine).
This commit fixes that by introducing a secondary ipset that is firstly
filled with the resolved IP addresses, and then its content is swapped
with the main set. This should hopefully reduce the likelihood of
random races.
We had this disabled because this was a rhel7 instance and couldn't run
linux-system-roles. Now it's 9, so re-enable.
Signed-off-by: Kevin Fenzi <kevin@scrye.com>
