Fixes
- The cluster autoscaler now picks up node config changes.
createupdated thecluster-autoscaler-configConfigMap and left the Deployment alone, and the autoscaler reads that config only at startup, so new autoscaled nodes kept the cloud-init from whenever the autoscaler had last started. The pod template now carrieshetzner-k3s/cluster-config-checksum, a sha256 of the config, so the Deployment rolls whenever the config changes: packages, pre and post k3s commands, the SSH port, allowed networks and the join address. - by @michal-billtech - Cloud-init gzip blocks now carry a fixed header timestamp, so an unchanged config produces identical bytes. Without it the checksum changed on every run and the autoscaler would have restarted on every
create. - by @michal-billtech
Upgrading
- Run
createonce after upgrading. The autoscaler rolls a single time, because the checksum annotation is new and the compressed cloud-init bytes changed. Masters and workers printNo change detectedand do not restart, and running pods are not affected. A secondcreatewith an unchanged config restarts nothing. Both were observed on a real three master cluster, and the config JSON was also rendered twice in a probe to confirm the checksum stays stable. - No bundled addon pins moved in this release: the cloud controller manager, the CSI driver, the autoscaler image and the system upgrade controller are at the same versions as 2.7.0.
If hetzner-k3s is useful to you or your company, please consider sponsoring its development.
Sponsorship helps ensure continued maintenance and new features. Thank you to our current sponsors! 🙏