---
# playbooks/proxmox/grow_vm_disks.yml
#
# Purpose:
#   Idempotently ensures all Swarm VM disks are sized to vm_disk_target on the
#   Proxmox layer (Play 1), reboots affected VMs so the guest kernel reads the
#   new block device geometry (Play 2), then grows the in-guest partition and
#   filesystem to match (Play 3).
#
# Architecture:
#   Play 1 — proxmox_cluster: checks the actual LVM volume size via `lvs` (NOT
#     `qm config`, which can be out of sync) and uses `lvextend` if below target.
#     WHY lvs not qm config: qm resize updates Proxmox metadata but can silently
#     fail to grow the LVM when the VM is running. lvs shows ground truth.
#   Play 2 — proxmox_cluster: reboots only the VMs whose LVs were just extended.
#   Play 3 — swarm_hosts: waits for SSH, then runs disk_grow role (growpart +
#     resize2fs). WHY reboot required: virtio-scsi guests on this kernel do not
#     honour /sys/class/block/sda/device/rescan or scsi_host scans while running.
#     Only a cold re-read of block device geometry at boot is reliable.
#
# VMID scheme: manager = (node_index * 100) + 1, worker = (node_index * 100) + 2
#   pve01 → 101/102, pve02 → 201/202, pve03 → 301/302
# LV path: /dev/pve/vm-{vmid}-disk-0  (standard local-lvm layout)
#
# Pre-requisites:
#   - SSH access to proxmox_cluster and swarm_hosts
#   - LVM tools available on Proxmox nodes (standard PVE install)
#   - cloud-guest-utils will be installed by disk_grow role if absent
#
# Usage:
#   Fix all Swarm VMs across all PVE nodes:
#     ansible-playbook -i inventory/hosts.ini playbooks/proxmox/grow_vm_disks.yml
#
#   Fix a single node end-to-end (all three plays, one guest):
#     ansible-playbook -i inventory/hosts.ini playbooks/proxmox/grow_vm_disks.yml \
#       -e "target_vmids=101" --limit pve01   # Play 1+2 on pve01, Play 3 on swarm-manager-1
#
#   In-guest grow only (disk already extended, VM already rebooted):
#     ansible-playbook -i inventory/hosts.ini playbooks/proxmox/grow_vm_disks.yml \
#       --limit swarm-manager-1 --tags in_guest
#
#   Validate only (no changes):
#     ansible-playbook -i inventory/hosts.ini playbooks/proxmox/grow_vm_disks.yml \
#       --check
#
# Verification after run:
#   ansible swarm_hosts -i inventory/hosts.ini -m shell -a "df -h /" --become

# ============================================================
# PLAY 1: Proxmox layer — extend LVM volume for each Swarm VM
# Source of truth: lvs (actual LVM size), NOT qm config (metadata only)
# ============================================================
- name: Extend Swarm VM LVM volumes on Proxmox hosts
  hosts: proxmox_cluster
  become: false
  gather_facts: false
  tags: [proxmox_resize]

  vars:
    vm_disk_target: "32G"
    vm_disk_target_gb: "{{ vm_disk_target | regex_replace('[^0-9]', '') | int }}"
    vm_lv_vg: "pve"

  tasks:
    - name: Derive VM IDs and LV names for this PVE node
      ansible.builtin.set_fact:
        disk_grow_manager_vmid: "{{ (inventory_hostname | regex_replace('[^0-9]', '') | int) * 100 + 1 }}"
        disk_grow_worker_vmid: "{{ (inventory_hostname | regex_replace('[^0-9]', '') | int) * 100 + 2 }}"

    # Manager VM -------------------------------------------------------
    # WHY lvs not qm config: qm resize updates Proxmox metadata but silently
    # fails to grow the LVM when the VM is already running. lvs is ground truth.

    - name: Get actual LVM size for manager VM {{ disk_grow_manager_vmid }}
      ansible.builtin.shell: |
        lvs --noheadings --units g -o lv_size \
          /dev/{{ vm_lv_vg }}/vm-{{ disk_grow_manager_vmid }}-disk-0 2>/dev/null \
          | tr -d ' ' | sed 's/g$//' | cut -d. -f1 \
          || echo "absent"
      args:
        executable: /bin/bash
      register: disk_grow_manager_lv_size
      changed_when: false

    - name: Extend manager VM LV to {{ vm_disk_target }} if below target
      ansible.builtin.shell: |
        lvextend -L {{ vm_disk_target }} \
          /dev/{{ vm_lv_vg }}/vm-{{ disk_grow_manager_vmid }}-disk-0
      args:
        executable: /bin/bash
      when:
        - disk_grow_manager_lv_size.stdout | trim != 'absent'
        - (disk_grow_manager_lv_size.stdout | trim | int) < (vm_disk_target_gb | int)
      register: disk_grow_manager_extend_result
      changed_when: disk_grow_manager_extend_result.rc == 0

    - name: Report manager VM LV state
      ansible.builtin.debug:
        msg: >-
          Manager VM {{ disk_grow_manager_vmid }} LV:
          {{ disk_grow_manager_lv_size.stdout | trim }}G
          → {{ vm_disk_target }}
          ({{ 'extended — reboot required' if (disk_grow_manager_extend_result is not skipped)
              else 'already at target or absent' }})
      when: disk_grow_manager_lv_size.stdout | trim != 'absent'

    # Worker VM --------------------------------------------------------

    - name: Get actual LVM size for worker VM {{ disk_grow_worker_vmid }}
      ansible.builtin.shell: |
        lvs --noheadings --units g -o lv_size \
          /dev/{{ vm_lv_vg }}/vm-{{ disk_grow_worker_vmid }}-disk-0 2>/dev/null \
          | tr -d ' ' | sed 's/g$//' | cut -d. -f1 \
          || echo "absent"
      args:
        executable: /bin/bash
      register: disk_grow_worker_lv_size
      changed_when: false

    - name: Extend worker VM LV to {{ vm_disk_target }} if below target
      ansible.builtin.shell: |
        lvextend -L {{ vm_disk_target }} \
          /dev/{{ vm_lv_vg }}/vm-{{ disk_grow_worker_vmid }}-disk-0
      args:
        executable: /bin/bash
      when:
        - disk_grow_worker_lv_size.stdout | trim != 'absent'
        - (disk_grow_worker_lv_size.stdout | trim | int) < (vm_disk_target_gb | int)
      register: disk_grow_worker_extend_result
      changed_when: disk_grow_worker_extend_result.rc == 0

    - name: Report worker VM LV state
      ansible.builtin.debug:
        msg: >-
          Worker VM {{ disk_grow_worker_vmid }} LV:
          {{ disk_grow_worker_lv_size.stdout | trim }}G
          → {{ vm_disk_target }}
          ({{ 'extended — reboot required' if (disk_grow_worker_extend_result is not skipped)
              else 'already at target or absent' }})
      when: disk_grow_worker_lv_size.stdout | trim != 'absent'

    # Reboot any VMs whose LV was just extended ---------------------------
    # WHY here not in Play 2: qm reboot runs on the PVE host, not the guest.
    # We only reboot VMs that were actually extended this run.

    - name: Reboot manager VM {{ disk_grow_manager_vmid }} to expose new disk size to guest kernel
      ansible.builtin.shell: qm reboot {{ disk_grow_manager_vmid }}
      args:
        executable: /bin/bash
      when:
        - disk_grow_manager_extend_result is not skipped
        - disk_grow_manager_extend_result.changed
      changed_when: true

    - name: Reboot worker VM {{ disk_grow_worker_vmid }} to expose new disk size to guest kernel
      ansible.builtin.shell: qm reboot {{ disk_grow_worker_vmid }}
      args:
        executable: /bin/bash
      when:
        - disk_grow_worker_extend_result is not skipped
        - disk_grow_worker_extend_result.changed
      changed_when: true

# ============================================================
# PLAY 2: Wait for rebooted Swarm nodes to come back
# ============================================================
- name: Wait for Swarm nodes to return after reboot
  hosts: swarm_hosts
  become: false
  gather_facts: false
  tags: [proxmox_resize, in_guest]

  tasks:
    - name: Wait for SSH to become available (up to 2 minutes)
      ansible.builtin.wait_for_connection:
        delay: 10
        timeout: 120

# ============================================================
# PLAY 3: In-guest layer — grow partition and filesystem
# ============================================================
- name: Grow in-guest root partition and filesystem on all Swarm nodes
  hosts: swarm_hosts
  become: true
  gather_facts: true
  tags: [in_guest]
  roles:
    - disk_grow