So, I have a client that has an internal use application that needs an
ancient version of libc5. That's not a typo; libc5. Before the server
that ran it died about a year and a half ago (said server was an AMD
K6-2/450 with a 6GB Western Digital Caviar drive that had been spinning
nearly continuously for almost 20 years!) it was running on Red Hat
Linux 5.2. The last version of CentOS that shipped with a libc5 was
2.1. So I set them up a KVM guest running CentOS 2.1, mainly because
Red Hat 5.2 wouldn't recognize the network at all, using the e1000
network driver, and it ran well.
Now, yes, there are no updates, but no it doesn't matter; it's an
internal use application that has a very small footprint and very low risk.
After the update on the host to 7.7.1908, the network stopped running.
The host also has a CentOS 7 guest that is still working properly. If I
change the 2.1 system to not automatically load the e1000 driver and
console in and 'modprobe e1000' manually, it starts working again, for a
while.
Any ideas as to where to start looking? qemu-kvm-ev itself, or
libvirt? I'm planning to start out with rollbacks to the previous
versions of each to try to find where the issue starts.