Fixes
- Reinstalling an external worker in place no longer leaves
k3s-agentstopped. When the k3s installer finds the regenerated service identical to the existing one it skips its own start; the worker install script now starts an inactive or failed unit itself and leaves a running agent alone. (#804) - by @fcavalieri - A first start that systemd restarts after the flannel annotation race is no longer reported as a failed install. The script waits while the unit is activating, fails at once for an inactive or failed unit, and prints the last 30 lines of the agent journal when it gives up. (#804) - by @fcavalieri
- The in-place reinstall instructions now live in the general external node setup section, and the worker install spec covers a failed explicit start of a stopped agent.
Upgrading
No action is required. The change is in the worker install script, so it takes effect the next time create runs or a worker is installed.
Healthy agents are not restarted: the k3s installer sees the same binary, unit file and environment file and skips its own service start, and the script only starts a unit that is inactive or failed. Existing external nodes are skipped while /etc/initialized is true and k3s-agent is active. Masters use a separate install script and are unchanged. No configuration keys change, so there is no config format bump.
No bundled addon pins changed in this release: the CCM, CSI, system upgrade controller and cluster autoscaler versions are the same as in v2.9.0.
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! 🙏