Upgrade kube-hetzner from 2.16.1 to 2.19.2

[s-schmidt-sl]

Hi,

I'm trying to upgrade the kube-hetzner module from 2.16.1 to current (2.19.2)
What would you recommend as upgrade step/path?

As of the release notes I might run into

• 2.17.1 placement groups-
• DO NOT Use 2.18.0
• 2.19.X - the 2.18 NAT router thing Which I do not have yet or plan to enable

I have test smaller cluster to tried different version steps like
2.16.1 >2.17.4
but no really success without down time.

My current solution, but it looses ingress, I have traefik pod on multiple nodes, but I loose the... LB connections , I guess, because of the internal network changes, which somehow rerenames eth1 back to en7s0 and I have to run the /etc/cloud/rename_interface.sh and reboot.

Not sure yet if k3s agent restart would work too
Any thoughts or advice to do it better w/o losing ingress?

here my script

sed -i 's/2.16.1/2.19.2/' kube.tf
grep version kube.tf
echo "press Enter";read
#terraform import -no-color 'module.kube-hetzner.hcloud_load_balancer_network.cluster[0]' '5871257-11970697'
tfi -upgrade
echo "press Enter";read
tfp
echo "press Enter";read
CLUSTER=$(hcloud load-balancer describe ststest-traefik --output json | jq .id)
NETWORK=$(hcloud network describe ststest --output json | jq .id)
terraform import -no-color 'module.kube-hetzner.hcloud_load_balancer_network.cluster[0]' "${CLUSTER}-${NETWORK}"


# update Ansible inventory (just in case)

echo -e "all:\n  hosts:">ansible/inventory.yml;hcloud server list -o json | \
    jq -r '.[] |"    " +.name  +":\n      ansible_host:  " +.public_net.ipv4.ip'>> ansible/inventory.yml
tfa -auto-approve& ## will break internal IP networks

# API gone
#### Ingress gone :( ####

sleep 60
# run rename and reboot on control plans
ansible-playbook ansible/rename-interface.yml -i ansible/inventory.yml --limit \*-control-plane\*&
sleep 60
tfa -auto-approve &

# run ansible on workers  2 at a time, customize the --limit like specific pool names

ansible-playbook ansible/rename-interface.yml -i ansible/inventory.yml \
-e serial_batch_size=2 \
--limit \*pool\*

[abriginets]

Might be a little late, but could you please take a look at this? terraform apply renamed eth1 to enp7s0 specifically. The udev rule approach is more robust than running rename_interface.sh repeatedly because it survives future reboots.

[s-schmidt-sl]

thx for the answer,
Now I actually have to do it since

What should you do?
Please update your hcloud-cloud-controller-manager to the latest available version, or at least to v1.30.1, to ensure uninterrupted > operation starting 1 July 2026.