GitLab runner update - fix samba-fips on Rackspace runners

[Stefan Metzmacher]

Hi Andrew,

> The old kernels were  4.15.0-159-generic from Ubuntu 18.04
> 
> The Kernel under our CI images is now 5.15.0-41-generic from Ubuntu
> 22.04

Can we use the hwe kernels there (linux-generic-hwe-22.04),
it would mean we would get 6.2.0-31-generic, and even more
recent ones once ubuntu 23.10 get stable.

It would mean we could do io_uring tests with the latest
zero copy features available with kernels >= 6.1

> The shared runners are 5.4.109+  BTW.
> 
> This seems to fix a persistent failure in the samba-fips test, eg as
> seen here: https://gitlab.com/samba-team/devel/samba/-/jobs/4840635606
> 
> I've also taken the opportunity to set the maximum job timeout in the
> Rackspace side to 4 hours - within 5 hours a VM will be terminated by
> cloud-checker.py, as script running on the bastion host, and to make it
> not block.  (Blocking deleting hosts is bad, sometimes the calls stall
> and fail to complete, preventing other VMs from being killed off).
> This should help keep our CI bill under control.
> 
> Also of interest is a MR to only use shared runners
> https://gitlab.com/samba-team/samba/-/merge_requests/3255
> 
> I'm shocked to see this all green!
> 
> This would mean we could save our private runners for security work
> only.
> 
> Finally, I've noticed that GitLab upstream has removed their reliance
> on Docker Hub, so the limitation (an aggressive rate limit) that
> blocked our move to OSU OSL a while back no longer applies.  If we
> chose to go the other direction (move more CI to OSU OSL, say if GitLab
> changes their CI terms/fees) that should now be entirely practical.

Thanks for all the work!

metze