[Bug]: kured wrong sentinel path about terraform-hcloud-kube-hetzner HOT 5 CLOSED

@jr-dimedis Are you using MicroOS nodes outside of this project? If so, yes they do behave differently. However we configure the nodes to work with the kured reboot method.

I will have a look again at your issue, but it would surprise me if there's really a problem there. Please do make sure to upgrade the project to the latest version of kube-hetzner.

from terraform-hcloud-kube-hetzner.

If you cat /etc/transactional-update.conf it should see REBOOT_METHOD=kured, this is what ensures that the file created is reboot-required and not reboot-needed.

from terraform-hcloud-kube-hetzner.

@mysticaltech

Thanks for your reply.

I created all nodes using kube-hetzner. The original nodes were created with an image from May 22, 2023, later some were created with an image from November 23, 2023. I started with kube-hetzner v2.1.6, then upgraded to v2.7.2 and later to v2.11.7.

Yes, the /etc/transactional-update.conf file contains REBOOT_METHOD=kured, but it looks that this config isn't used when you just execute transactional-update in the shell. The service file adds this config file using environment variables, that's why it should be triggered via the systemd service. I checked this with and without using systemctl and I was able to clearly reproduce the behavior in this way.

# That's the systemd service of a node updated Feb 03
$ cat /usr/lib/systemd/system/transactional-update.service 
[Unit]
Description=Update the system
Documentation=man:transactional-update(8)
ConditionACPower=true
Wants=network.target
After=network.target

[Service]
Type=oneshot
Environment=UPDATE_METHOD=dup
EnvironmentFile=-/usr/etc/transactional-update.conf
EnvironmentFile=-/etc/transactional-update.conf
ExecStart=/usr/sbin/transactional-update cleanup ${UPDATE_METHOD} reboot
IOSchedulingClass=best-effort
IOSchedulingPriority=7

What about the reboot-sentinel option?

The Pod mounts the hosts /var/run folder to /sentinel and for me it only works, when I change the path in the configuration (here an excerpt from a running Pod yaml):

apiVersion: v1
kind: Pod
metadata:
  name: kured-94hjw
  namespace: kube-system
  ...
...
spec:
  containers:
    ...
    volumeMounts:
    - mountPath: /sentinel
      name: sentinel
      readOnly: true

  ...
  volumes:
  - hostPath:
      path: /var/run
      type: Directory
    name: sentinel
...

And yes, I'll try the latest version, even though I haven't seen any changes in that area - but of course I could be wrong ;)

from terraform-hcloud-kube-hetzner.

I just upgraded kube-hetzner on a test cluster to latest v2.11.8. No changes on the cluster, in particular no Kured redeployment or something like that. So this version bump makes no difference.

from terraform-hcloud-kube-hetzner.

@jr-dimedis I will investigate this more, thanks for the info. Keep you posted.

from terraform-hcloud-kube-hetzner.