How often do you actually reboot for kernel patches?

He literally released 6.11-ck1 last month. The AUR package is a convenience wrapper. I build from source with my own PKGBUILD because I don't trust binaries. This is basic.

Ubuntu is a rolling release with extra steps and worse testing. Debian stable plus backports if you need enterprise, or Arch if you have a spine. The livepatch kernel is a different ABI than the stock kernel and that is how you get weird module load failures at 2 AM.

My RackNerd NAT box: $12/year, 847 days, runs custom kernel, no livepatch, no reboot. The graphs don't lie.

I stand corrected. One release in 18 months. Very current. Very supported.

The ABI thing is real though. Canonical's livepatch kernel is a separate build and I've seen WireGuard kmod fail to load after a "seamless" patch. Not often. Once is enough when it's production.

Oracle free tier terms say you can't use it for CI. They don't enforce it until they do, and then they terminate with 24 hours notice. Good luck.

Following this thread with interest. Small shop, 8 VPS across 3 providers, no livepatch, reboot on apt upgrade which is weekly. Never had a problem that wasn't self-inflicted.

Question: does anyone actually test their livepatch before it hits production? Like, staging environment with the same workload? Or do you just trust Canonical?

No. The whole point is it's automatic. If I wanted to test kernel patches I'd run my own kernel team.

What I do test: the reboot path. Monthly I pick one node and reboot it during business hours to verify services come back clean. Caught a systemd unit that failed to start after network-online.target changed behavior. Livepatch can't save you from userspace.

This. A thousand times this. The reboot is the only integration test that matters. You find the fstab with the dead UUID, the NFS mount that hangs forever, the custom sysctl that got overwritten by a package update. Livepatch hides rot.

I don't hate livepatch as a concept. I hate what it enables: the fantasy that you never need to verify your system's actual boot path.

Same here on the reboot testing. I do it quarterly and I've found issues every single time. Last one: mdadm array started degraded because one disk was slow to respond and the new kernel had a changed timeout. Would livepatch have caught that? No.

Also: has anyone else noticed Contabo's St. Louis routing to Level3 got worse in August? Used to be 35ms to Dallas, now 55ms.