[markuslf]

Problem

On Rocky 8 and 9, cloud-init's sysconfig renderer writes dns = none to /etc/NetworkManager/conf.d/99-cloud-init.conf whenever the provider hands out DNS servers (Hetzner does). The cloud_init role removes the file but never reloads NetworkManager, so /etc/resolv.conf stays unmanaged until the next NetworkManager restart or reboot. Rocky 10 lists network-manager before sysconfig in cloud.cfg and does not write the file.

Change

• roles/cloud_init: removing the file notifies a handler running systemctl try-reload-or-restart NetworkManager.service.
  • ansible.builtin.service with state: reloaded would start a stopped NetworkManager (ansible-core 2.16, systemd_service.py); try-reload-or-restart leaves it alone. NetworkManager has an ExecReload, so this never restarts.
  • A reload re-reads conf.d and re-evaluates the DNS mode without touching connections (NetworkManager 1.56, src/core/dns/nm-dns-manager.c, config_changed_cb).
  • rc 5 (unit not found) is tolerated for hosts without NetworkManager.
• README: the description of the file was inverted; Tags list the reload as trigger.
• setup_basic Molecule scenario: prepare.yml recreates the leftover file on every host running NetworkManager and asserts the DNS mode is none; verify.yml asserts it is no longer none.
• CHANGELOG updated.

Testing

setup_basic on all 8 targets, in two batches (idempotence before verify):

• Debian 12/13, Ubuntu 22.04/24.04/26.04: converge, idempotence (changed=0) and verify green. NetworkManager does not run there, so the new steps are skipped.
• Rocky 8/9/10: converge green, handler fires, idempotence changed=0, new assertion green.
• Negative check: with the previous role code, the new assertion fails on Rocky 8/9/10 (DNS mode stays none).

The remaining verify failures on Rocky are unrelated and come from the linuxfabrik-monitoring-plugins 8.0.0 RPM: a corrupt .pyc breaks about-me on Rocky 9, and AIDE reports rewritten .pyc files on Rocky 8 and 10 (same as noted in #453).