- name: Kubelet image garbage collection
  tags: kubelet_gc
  block:
    - name: Create the kubelet drop-in directory
      file:
        path: /etc/rancher/k3s/kubelet.conf.d
        state: directory
        mode: "0755"

    - name: Set the image garbage collection thresholds
      copy:
        dest: /etc/rancher/k3s/kubelet.conf.d/10-image-gc.conf
        mode: "0644"
        content: |
          apiVersion: kubelet.config.k8s.io/v1beta1
          kind: KubeletConfiguration
          imageGCHighThresholdPercent: 70
          imageGCLowThresholdPercent: 55
          imageMaximumGCAge: 168h

# KubeletConfiguration fields from kubernetes.kubelet_memory (eviction) and
# kubernetes.kubelet_reserved (systemReserved, kubeReserved); a file is removed
# while its variable is empty. The kubelet reads the drop-ins only when k3s
# starts, so a change takes effect after a k3s restart. The restart_k3s handler
# restarts only with -e k3s_restart_allowed=true, on one node per run.
- name: Kubelet memory reservation and eviction
  tags: [kubelet_memory, kubelet_reserved]
  block:
    - name: Refuse a k3s restart on several nodes at once
      ansible.builtin.assert:
        that: ansible_play_hosts_all | length == 1
        fail_msg: k3s_restart_allowed restarts k3s on every changed node of the play; limit the run to one node (--limit <node>) to keep the etcd quorum
      when: k3s_restart_allowed | default(false) | bool

    - name: Create the kubelet drop-in directory
      file:
        path: /etc/rancher/k3s/kubelet.conf.d
        state: directory
        mode: "0755"

    - name: Set the kubelet eviction thresholds
      copy:
        dest: /etc/rancher/k3s/kubelet.conf.d/20-memory.conf
        mode: "0644"
        content: "{{ {'apiVersion': 'kubelet.config.k8s.io/v1beta1', 'kind': 'KubeletConfiguration'} | combine(kubernetes.kubelet_memory) | to_nice_yaml(indent=2) }}"
      when: kubernetes.kubelet_memory | length > 0
      register: kubelet_memory_dropin
      notify: restart_k3s

    - name: Remove the kubelet memory drop-in
      file:
        path: /etc/rancher/k3s/kubelet.conf.d/20-memory.conf
        state: absent
      when: kubernetes.kubelet_memory | length == 0
      notify: restart_k3s

    - name: Set the kubelet system and kube reservations
      copy:
        dest: /etc/rancher/k3s/kubelet.conf.d/30-reserved.conf
        mode: "0644"
        content: "{{ {'apiVersion': 'kubelet.config.k8s.io/v1beta1', 'kind': 'KubeletConfiguration'} | combine(kubernetes.kubelet_reserved) | to_nice_yaml(indent=2) }}"
      when: kubernetes.kubelet_reserved | length > 0
      register: kubelet_reserved_dropin
      notify: restart_k3s

    - name: Remove the kubelet reservation drop-in
      file:
        path: /etc/rancher/k3s/kubelet.conf.d/30-reserved.conf
        state: absent
      when: kubernetes.kubelet_reserved | length == 0
      notify: restart_k3s

    # Compares the file times with the start of the running k3s, so a pending
    # restart is reported on every run, not only on the run that wrote the file.
    - name: List the kubelet drop-ins written after k3s started
      shell: |
        since=$(systemctl show k3s -p ActiveEnterTimestamp --value)
        [ -n "$since" ] || exit 0
        start=$(date -d "$since" +%s) || exit 1
        for f in /etc/rancher/k3s/kubelet.conf.d/*; do
          [ "$(stat -c %Y "$f")" -gt "$start" ] && echo "$f"
        done
        exit 0
      register: kubelet_dropins_pending
      changed_when: false
      check_mode: false

    - name: Report the pending k3s restart
      ansible.builtin.debug:
        msg: >-
          k3s restart pending on {{ inventory_hostname }}: the kubelet reads
          /etc/rancher/k3s/kubelet.conf.d only at start. Restart one node at a
          time with --limit {{ inventory_hostname }} -e k3s_restart_allowed=true.
      when:
        - not (k3s_restart_allowed | default(false) | bool)
        - kubelet_dropins_pending.stdout | length > 0 or kubelet_memory_dropin is changed or kubelet_reserved_dropin is changed

# Every pod watching its config files takes inotify instances and watches from
# the per-user pool. An LXC container inherits these from its host. Only this
# file is loaded: sysctl --system would also apply distribution defaults such as
# rp_filter that differ from the live values Cilium relies on.
- name: Raise the inotify limits
  tags: inotify
  when: ansible_facts['virtualization_type'] != 'lxc'
  ansible.posix.sysctl:
    name: "{{ item.key }}"
    value: "{{ item.value }}"
    sysctl_file: /etc/sysctl.d/90-inotify.conf
  loop: "{{ {'fs.inotify.max_user_instances': 65536, 'fs.inotify.max_user_watches': 4194304} | dict2items }}"