ansible-msp-automations/roles/proxmox_upgrade/tasks/drain.yml

---
# =============================================================================
# proxmox_upgrade — drain.yml
# Migrate all VMs/LXCs off current_node before upgrading
# KVM: community.proxmox.proxmox_kvm (API, delegate_to: localhost)
# LXC: pct migrate (SSH on source node)
# =============================================================================

# ── Discover guests on this node ──────────────────────────────────────────────
- name: "Drain | Get all guests on {{ current_node }}"
  community.proxmox.proxmox_vm_info:
    api_host: "{{ api_host }}"
    api_user: "{{ api_user }}"
    api_token_id: "{{ api_token_id }}"
    api_token_secret: "{{ api_token_secret }}"
    api_port: "{{ api_port }}"
    node: "{{ current_node }}"
  register: node_guests
  delegate_to: localhost

- name: "Drain | Get available target nodes"
  community.proxmox.proxmox_node_info:
    api_host: "{{ api_host }}"
    api_user: "{{ api_user }}"
    api_token_id: "{{ api_token_id }}"
    api_token_secret: "{{ api_token_secret }}"
    api_port: "{{ api_port }}"
  register: all_nodes_info
  delegate_to: localhost

- name: "Drain | Set migration target"
  ansible.builtin.set_fact:
    migration_target: >-
      {{ all_nodes_info.proxmox_nodes
         | selectattr('status', 'equalto', 'online')
         | rejectattr('node', 'equalto', current_node)
         | map(attribute='node')
         | list
         | first }}
  delegate_to: localhost

- name: "Drain | Fail if no migration target available"
  ansible.builtin.fail:
    msg: "No online nodes available to migrate guests to. Cannot drain {{ current_node }}."
  when: migration_target is not defined or migration_target == ''
  delegate_to: localhost

- name: "Drain | Log raw guest data (for debugging)"
  ansible.builtin.debug:
    msg: >-
      Guests on {{ current_node }}:
      vmids={{ node_guests.proxmox_vms | map(attribute='vmid') | list }}
      first_vm_keys={{ node_guests.proxmox_vms | first | list }}
  delegate_to: localhost

- name: "Drain | Build KVM migration list"
  ansible.builtin.set_fact:
    kvm_guests: >-
      {{ node_guests.proxmox_vms
         | selectattr('type', 'equalto', 'qemu')
         | list }}
  delegate_to: localhost

- name: "Drain | Build LXC migration list"
  ansible.builtin.set_fact:
    lxc_guests: >-
      {{ node_guests.proxmox_vms
         | selectattr('type', 'equalto', 'lxc')
         | list }}
  delegate_to: localhost

- name: "Drain | Log migration plan"
  ansible.builtin.debug:
    msg: >-
      Drain plan for {{ current_node }} → {{ migration_target }}:
      KVM ({{ kvm_guests | length }}): {{ kvm_guests | map(attribute='vmid') | list }}
      LXC ({{ lxc_guests | length }}): {{ lxc_guests | map(attribute='vmid') | list }}
  delegate_to: localhost

# ── KVM migrations ────────────────────────────────────────────────────────────
- name: "Drain | Migrate KVM guests"
  when: kvm_guests | length > 0
  block:
    - name: "Drain | KVM | Live migrate (sequential)"
      community.proxmox.proxmox_kvm:
        api_host: "{{ api_host }}"
        api_user: "{{ api_user }}"
        api_token_id: "{{ api_token_id }}"
        api_token_secret: "{{ api_token_secret }}"
        api_port: "{{ api_port }}"
        node: "{{ current_node }}"
        vmid: "{{ item.vmid }}"
        migrate: true
        target: "{{ migration_target }}"
        timeout: "{{ vm_shutdown_timeout }}"
      loop: "{{ kvm_guests }}"
      delegate_to: localhost
      when: not migration_bulk | bool

    - name: "Drain | KVM | Bulk migrate (fire all simultaneously)"
      community.proxmox.proxmox_kvm:
        api_host: "{{ api_host }}"
        api_user: "{{ api_user }}"
        api_token_id: "{{ api_token_id }}"
        api_token_secret: "{{ api_token_secret }}"
        api_port: "{{ api_port }}"
        node: "{{ current_node }}"
        vmid: "{{ item.vmid }}"
        migrate: true
        target: "{{ migration_target }}"
        timeout: "{{ vm_shutdown_timeout }}"
      loop: "{{ kvm_guests }}"
      delegate_to: localhost
      async: "{{ vm_shutdown_timeout * 2 }}"
      poll: 0
      register: kvm_bulk_jobs
      when: migration_bulk | bool

    - name: "Drain | KVM | Wait for bulk migrations to complete"
      ansible.builtin.async_status:
        jid: "{{ item.ansible_job_id }}"
      register: kvm_job_result
      until: kvm_job_result.finished
      retries: 60
      delay: 10
      loop: "{{ kvm_bulk_jobs.results }}"
      delegate_to: localhost
      when: migration_bulk | bool

# ── LXC migrations ────────────────────────────────────────────────────────────
- name: "Drain | Migrate LXC guests"
  when: lxc_guests | length > 0
  block:
    - name: "Drain | LXC | Warn about restart requirement"
      ansible.builtin.debug:
        msg: >-
          LXC {{ item.vmid }} ({{ item.name | default('unknown') }}) will be
          stopped, migrated to {{ migration_target }}, and restarted.
          (LXC live migration is not supported by Proxmox)
      loop: "{{ lxc_guests | selectattr('status', 'equalto', 'running') | list }}"
      delegate_to: localhost

    - name: "Drain | LXC | Skip warning"
      ansible.builtin.debug:
        msg: >-
          WARNING — LXC {{ item.vmid }} ({{ item.name | default('unknown') }})
          live_migrate_fallback=skip — THIS CONTAINER WILL GO DOWN during node reboot.
      loop: "{{ lxc_guests | selectattr('status', 'equalto', 'running') | list }}"
      when: live_migrate_fallback == 'skip'
      delegate_to: localhost

    - name: "Drain | LXC | Migrate via pct migrate"
      ansible.builtin.command: >
        pct migrate {{ item.vmid }} {{ migration_target }}
        {% if item.status == 'running' %}--restart{% endif %}
        --timeout {{ lxc_migrate_timeout }}
      loop: "{{ lxc_guests }}"
      when: live_migrate_fallback != 'skip'
      changed_when: true

    - name: "Drain | LXC | Log migration results"
      ansible.builtin.debug:
        msg: "LXC {{ item.item.vmid }} migrated to {{ migration_target }}"
      loop: "{{ lxc_migrate_result.results | default([]) }}"
      when: item.rc is defined and item.rc == 0

- name: "Drain | {{ current_node }} drained successfully"
  ansible.builtin.debug:
    msg: >-
      Node {{ current_node }} drained —
      {{ kvm_guests | length }} KVM + {{ lxc_guests | length }} LXC guests
      migrated to {{ migration_target }}