prole/infrastructure/playbooks/k3s_sync.yml

---
- name: Prepare k3s synchronization metadata
  hosts: k3s_hosts
  gather_facts: false
  run_once: true
  tasks:
    - name: Initialize k3s sync source host
      ansible.builtin.set_fact:
        k3s_sync_source_host: ""
      delegate_to: localhost
      delegate_facts: true

    - name: Select k3s init server as sync source
      ansible.builtin.set_fact:
        k3s_sync_source_host: "{{ item }}"
      loop: "{{ groups['k3s_hosts'] }}"
      when:
        - hostvars[item].k3s_role | default('agent') == 'server'
        - hostvars[item].k3s_cluster_init | default(false) | bool
      delegate_to: localhost
      delegate_facts: true

    - name: Fallback to first k3s server as sync source
      ansible.builtin.set_fact:
        k3s_sync_source_host: "{{ item }}"
      loop: "{{ groups['k3s_hosts'] }}"
      when:
        - (hostvars['localhost'].k3s_sync_source_host | length) == 0
        - hostvars[item].k3s_role | default('agent') == 'server'
      delegate_to: localhost
      delegate_facts: true

    - name: Require a k3s server as sync source
      ansible.builtin.assert:
        that:
          - hostvars['localhost'].k3s_sync_source_host | length > 0
        fail_msg: >-
          No k3s sync source found. Ensure at least one host has `k3s_role: server`.
          (If you have multiple servers and want to prefer the bootstrap server,
          set `k3s_cluster_init: true` on exactly one server.)          

    - name: Ensure sync source is included in this run
      ansible.builtin.assert:
        that:
          - hostvars['localhost'].k3s_sync_source_host in ansible_play_hosts_all
        fail_msg: "Sync source {{ hostvars['localhost'].k3s_sync_source_host }} is not in this run. Include it in --limit."

    - name: Create local temp directory for k3s sync
      ansible.builtin.command: mktemp -d -p /tmp k3s-sync-XXXXXX
      register: k3s_sync_tmpdir
      changed_when: true
      delegate_to: localhost
      become: false

    - name: Ensure k3s sync temp directory is writable
      ansible.builtin.file:
        path: "{{ k3s_sync_tmpdir.stdout }}"
        state: directory
        mode: "1777"
      delegate_to: localhost

    - name: Store k3s sync metadata on controller
      ansible.builtin.set_fact:
        k3s_sync_source_host: "{{ hostvars['localhost'].k3s_sync_source_host }}"
        k3s_sync_tmpdir_path: "{{ k3s_sync_tmpdir.stdout }}"
        k3s_sync_tls_bundle: "{{ k3s_sync_tmpdir.stdout }}/k3s-tls.tgz"
      delegate_to: localhost
      delegate_facts: true

- name: Collect k3s token and certs from init server
  hosts: k3s_hosts
  become: true
  gather_facts: false
  run_once: true
  vars:
    k3s_sync_source_host: "{{ hostvars['localhost'].k3s_sync_source_host | default('') }}"
    k3s_sync_tls_bundle: "{{ hostvars['localhost'].k3s_sync_tls_bundle | default('') }}"
  tasks:
    - name: Require a k3s sync source
      ansible.builtin.assert:
        that:
          - k3s_sync_source_host | length > 0
        fail_msg: "k3s sync source is empty. Ensure the init server is reachable."

    - name: Ensure k3s node token exists
      ansible.builtin.stat:
        path: "{{ k3s_data_dir }}/server/node-token"
      register: k3s_node_token
      delegate_to: "{{ k3s_sync_source_host }}"

    - name: Fail when k3s node token is missing
      ansible.builtin.fail:
        msg: "k3s node token not found at {{ k3s_data_dir }}/server/node-token"
      when: not k3s_node_token.stat.exists

    - name: Read k3s node token
      ansible.builtin.slurp:
        src: "{{ k3s_data_dir }}/server/node-token"
      register: k3s_node_token_raw
      delegate_to: "{{ k3s_sync_source_host }}"

    - name: Store k3s sync token on controller
      ansible.builtin.set_fact:
        k3s_sync_token: "{{ k3s_node_token_raw.content | b64decode | trim }}"
      delegate_to: localhost
      delegate_facts: true

    - name: Check for k3s tls directory
      ansible.builtin.stat:
        path: "{{ k3s_data_dir }}/server/tls"
      register: k3s_tls_dir
      delegate_to: "{{ k3s_sync_source_host }}"

    - name: Create k3s tls bundle
      ansible.builtin.archive:
        path: "{{ k3s_data_dir }}/server/tls"
        dest: /tmp/k3s-tls.tgz
        format: gz
      when: k3s_tls_dir.stat.exists
      delegate_to: "{{ k3s_sync_source_host }}"

    - name: Fetch k3s tls bundle
      ansible.builtin.fetch:
        src: /tmp/k3s-tls.tgz
        dest: "{{ k3s_sync_tls_bundle }}"
        flat: true
      when: k3s_tls_dir.stat.exists
      delegate_to: "{{ k3s_sync_source_host }}"

    - name: Mark tls bundle presence on controller
      ansible.builtin.set_fact:
        k3s_sync_tls_bundle_present: "{{ k3s_tls_dir.stat.exists }}"
      delegate_to: localhost
      delegate_facts: true

    - name: Remove temporary tls bundle from source
      ansible.builtin.file:
        path: /tmp/k3s-tls.tgz
        state: absent
      when: k3s_tls_dir.stat.exists
      delegate_to: "{{ k3s_sync_source_host }}"

- name: Synchronize k3s token and certs to servers
  hosts: k3s_hosts
  become: true
  serial: 1
  gather_facts: false
  vars:
    k3s_sync_source_host: "{{ hostvars['localhost'].k3s_sync_source_host }}"
    k3s_sync_token: "{{ hostvars['localhost'].k3s_sync_token | default('') }}"
    k3s_sync_tls_bundle: "{{ hostvars['localhost'].k3s_sync_tls_bundle | default('') }}"
    k3s_sync_tls_bundle_present: "{{ hostvars['localhost'].k3s_sync_tls_bundle_present | default(false) }}"
  pre_tasks:
    - name: Require k3s sync token
      ansible.builtin.assert:
        that:
          - k3s_sync_token | length > 0
        fail_msg: "k3s sync token is empty. Check the init server token."
  tasks:
    - name: Stop k3s before syncing
      ansible.builtin.import_role:
        name: k3s
        tasks_from: stop
      when: inventory_hostname != k3s_sync_source_host

    - name: Sync k3s token and certs
      ansible.builtin.import_role:
        name: k3s
        tasks_from: sync
      when: inventory_hostname != k3s_sync_source_host

    - name: Start k3s after syncing
      ansible.builtin.import_role:
        name: k3s
        tasks_from: start
      when: inventory_hostname != k3s_sync_source_host

- name: Update k3s vault token on controller
  hosts: k3s_hosts
  gather_facts: false
  run_once: true
  vars:
    k3s_sync_token: "{{ hostvars['localhost'].k3s_sync_token | default('') }}"
    vault_k3s_path: "{{ playbook_dir }}/../inventory/group_vars/all/vault_k3s.yml"
    vault_pass_default: "{{ playbook_dir }}/../../.vault_pass"
  tasks:
    - name: Skip vault update when disabled
      ansible.builtin.meta: end_play
      when: not (k3s_sync_update_vault | default(true) | bool)

    - name: Require k3s sync token for vault update
      ansible.builtin.assert:
        that:
          - k3s_sync_token | length > 0
        fail_msg: "k3s sync token is empty. Unable to update vault."

    - name: Check for default vault password file
      ansible.builtin.stat:
        path: "{{ vault_pass_default }}"
      register: vault_pass_default_stat
      delegate_to: localhost

    - name: Determine vault password file
      ansible.builtin.set_fact:
        k3s_vault_password_file: >-
          {{ k3s_vault_password_file
             | default(lookup('env', 'ANSIBLE_VAULT_PASSWORD_FILE') | default('', true), true) }}          
      delegate_to: localhost
      delegate_facts: true

    - name: Fallback to default vault password file
      ansible.builtin.set_fact:
        k3s_vault_password_file: "{{ vault_pass_default }}"
      when:
        - (hostvars['localhost'].k3s_vault_password_file | default('')) | length == 0
        - vault_pass_default_stat.stat.exists
      delegate_to: localhost
      delegate_facts: true

    - name: Require vault password file
      ansible.builtin.assert:
        that:
          - (hostvars['localhost'].k3s_vault_password_file | default('')) | length > 0
        fail_msg: "Set k3s_vault_password_file or ANSIBLE_VAULT_PASSWORD_FILE to update vault."

    - name: Ensure vault file exists
      ansible.builtin.stat:
        path: "{{ vault_k3s_path }}"
      register: vault_k3s_file
      delegate_to: localhost

    - name: Fail when vault file is missing
      ansible.builtin.fail:
        msg: "Vault file not found at {{ vault_k3s_path }}"
      when: not vault_k3s_file.stat.exists

    - name: Check if vault file is encrypted
      ansible.builtin.command: "head -n 1 {{ vault_k3s_path }}"
      register: vault_k3s_head
      changed_when: false
      delegate_to: localhost

    - name: Mark vault encryption state
      ansible.builtin.set_fact:
        vault_k3s_encrypted: "{{ (vault_k3s_head.stdout | default('')) is search('^\\$ANSIBLE_VAULT') }}"
      delegate_to: localhost
      delegate_facts: true

    - name: Decrypt vault_k3s.yml
      ansible.builtin.command: >-
        ansible-vault decrypt {{ vault_k3s_path }}
        --vault-password-file {{ hostvars['localhost'].k3s_vault_password_file }}        
      changed_when: true
      delegate_to: localhost
      when: hostvars['localhost'].vault_k3s_encrypted | default(false)

    - name: Update vault k3s token
      ansible.builtin.lineinfile:
        path: "{{ vault_k3s_path }}"
        regexp: '^vault_k3s_token:'
        line: "vault_k3s_token: \"{{ k3s_sync_token }}\""
      delegate_to: localhost

    - name: Encrypt vault_k3s.yml
      ansible.builtin.command: >-
        ansible-vault encrypt {{ vault_k3s_path }}
        --vault-password-file {{ hostvars['localhost'].k3s_vault_password_file }}        
      changed_when: true
      delegate_to: localhost
      when: hostvars['localhost'].vault_k3s_encrypted | default(false)

- name: Refresh kubeconfig on controller
  hosts: k3s_hosts
  gather_facts: false
  run_once: true
  tasks:
    - name: Fetch and save kubeconfig on controller
      ansible.builtin.import_role:
        name: k3s
        tasks_from: fetch_kubeconfig

- name: Validate k3s nodes from controller
  hosts: k3s_hosts
  gather_facts: false
  run_once: true
  vars:
    k3s_kubeconfig_service_path: "{{ playbook_dir }}/../../etc/secrets/k3s.kubeconfig"
    k3s_kubeconfig_project_path: "{{ playbook_dir }}/../../prole-k3s.kubeconfig"
    k3s_expected_node_count: "{{ groups['k3s_hosts'] | length }}"
  tasks:
    - name: Check for service kubeconfig
      ansible.builtin.stat:
        path: "{{ k3s_kubeconfig_service_path }}"
      register: _k3s_kubeconfig_service_stat
      delegate_to: localhost
      become: false

    - name: Check for project kubeconfig
      ansible.builtin.stat:
        path: "{{ k3s_kubeconfig_project_path }}"
      register: _k3s_kubeconfig_project_stat
      delegate_to: localhost
      become: false

    - name: Select kubeconfig path for validation
      ansible.builtin.set_fact:
        k3s_kubeconfig_validation_path: >-
          {{ k3s_kubeconfig_service_path
             if _k3s_kubeconfig_service_stat.stat.exists
             else k3s_kubeconfig_project_path }}          
      changed_when: false

    - name: Require a kubeconfig for validation
      ansible.builtin.assert:
        that:
          - _k3s_kubeconfig_service_stat.stat.exists or _k3s_kubeconfig_project_stat.stat.exists
        fail_msg: >-
          No kubeconfig found at {{ k3s_kubeconfig_service_path }} or {{ k3s_kubeconfig_project_path }}.
          Run `k3s_reset` or `k3s_sync` again and ensure the init server is reachable.          

    - name: Show nodes (kubectl get nodes -o wide)
      ansible.builtin.command: "kubectl --kubeconfig {{ k3s_kubeconfig_validation_path }} get nodes -o wide"
      register: _k3s_nodes_wide
      changed_when: false
      delegate_to: localhost
      become: false

    - name: Debug nodes output
      ansible.builtin.debug:
        var: _k3s_nodes_wide.stdout_lines

    - name: Count nodes
      ansible.builtin.shell: |
        set -euo pipefail
        kubectl --kubeconfig "{{ k3s_kubeconfig_validation_path }}" get nodes --no-headers | wc -l | tr -d ' '        
      args:
        executable: /bin/bash
      register: _k3s_node_count
      changed_when: false
      delegate_to: localhost
      become: false

    - name: Require expected node count
      ansible.builtin.assert:
        that:
          - (_k3s_node_count.stdout | int) >= (k3s_expected_node_count | int)
        fail_msg: >-
          Expected at least {{ k3s_expected_node_count }} nodes, but kubectl reported {{ _k3s_node_count.stdout | default('0') }}.
          Ensure all `k3s_hosts` are reachable and have rejoined, then re-run `k3s_sync`.