ansible/playbooks/nextcloud-upgrade.yml

- name: Resilient Nextcloud upgrade on Rocky Linux
  hosts: all
  become: yes
  vars:
    # --- Configuration Variables ---
    nextcloud_dir: /var/www/nextcloud
    web_user: apache
    php_cli: /usr/bin/php
    service_name: httpd

    # --- Environment for occ commands ---
    # This avoids setting TMPDIR on every single task
    occ_env:
      TMPDIR: /tmp

    # --- Core Apps ---
    # Add any app you consider "core" and do NOT want to disable.
    # user_ldap is a good example if you use it for auth.
    core_apps:
      - activity
      - twofactor_nextcloud_notification
      - files
      - settings
      - dav
      - updater
      - workflowengine
      - theming
      - user_ldap
      - dashboard
      - federation
      - updatestatus
      - accessibility
      - serverinfo

  tasks:

    - name: Reminder - ensure backups exist
      debug:
        msg: "⚠️ CRITICAL: Ensure you have valid DB + data directory backups BEFORE PROCEEDING!"
      
    - name: Pause playbook for 10 seconds to allow backup cancellation
      pause:
        seconds: 10
        prompt: "Press Ctrl+C and then 'A' to abort if you do not have backups."

    # ----------------------------------------------------------------------
    # Step 1: Check current version
    # ----------------------------------------------------------------------
    - name: Get current Nextcloud version
      command: "{{ php_cli }} occ status"
      args:
        chdir: "{{ nextcloud_dir }}"
      become_user: "{{ web_user }}"
      register: occ_status
      changed_when: false
      failed_when: "'installed: true' not in occ_status.stdout"

    - name: Parse and show current version
      set_fact:
        current_version: "{{ occ_status.stdout | regex_search('versionstring: (.*)', '\\1') | first | default('unknown') }}"
      when: occ_status.stdout is defined

    - name: Show current version
      debug:
        msg: "Current Nextcloud version: {{ current_version }}"

    # ----------------------------------------------------------------------
    # Step 2: Disable 3rd party apps
    # ----------------------------------------------------------------------
    - name: Get list of installed apps
      command: "{{ php_cli }} occ app:list"
      args:
        chdir: "{{ nextcloud_dir }}"
      become_user: "{{ web_user }}"
      register: app_list
      changed_when: false

    - name: Disable all non-core 3rd party apps
      command: "{{ php_cli }} occ app:disable {{ item }}"
      args:
        chdir: "{{ nextcloud_dir }}"
      become_user: "{{ web_user }}"
      environment: "{{ occ_env }}"
      loop: "{{ app_list.stdout_lines | select('match', '^[ ]*-[ ](.*):[ ]enabled$') | map('regex_replace', '^[ ]*-[ ](.*):[ ]enabled$', '\\1') | list }}"
      register: app_disable_result
      when: item not in core_apps
      ignore_errors: yes # Don't fail if an app can't be disabled

    - name: Save list of successfully disabled apps for re-enabling
      set_fact:
        disabled_apps: "{{ app_disable_result.results | selectattr('changed', 'defined') | selectattr('changed') | map(attribute='item') | list }}"
      when: app_disable_result is defined and app_disable_result.results is defined

    # ----------------------------------------------------------------------
    # Step 3: Begin Upgrade Transaction Block
    # ----------------------------------------------------------------------
    - name: Perform Nextcloud Upgrade (Transactional)
      block:

        - name: Check if maintenance mode is already active
          command: "{{ php_cli }} occ maintenance:mode"
          args:
            chdir: "{{ nextcloud_dir }}"
          become_user: "{{ web_user }}"
          register: maint_status
          changed_when: false
          failed_when: false # Don't fail, just check output

        - name: Enable maintenance mode
          command: "{{ php_cli }} occ maintenance:mode --on"
          args:
            chdir: "{{ nextcloud_dir }}"
          become_user: "{{ web_user }}"
          environment: "{{ occ_env }}"
          when: "'maintenance mode is not active' in maint_status.stdout"

        - name: Run Nextcloud CLI updater (with retry logic)
          command: "{{ php_cli }} updater/updater.phar --no-interaction"
          args:
            chdir: "{{ nextcloud_dir }}"
          become_user: "{{ web_user }}"
          environment: "{{ occ_env }}"
          register: updater_result
          until: 
            - updater_result.rc == 0
            - "'Update successful' in updater_result.stdout or 'Nextcloud is already latest version' in updater_result.stdout"
          retries: 3
          delay: 10
          # This is the final failure check for the updater itself
          failed_when: 
            - updater_result.rc != 0
            - "'Nextcloud is already latest version' not in updater_result.stdout"

        - name: Run occ upgrade
          command: "{{ php_cli }} occ upgrade"
          args:
            chdir: "{{ nextcloud_dir }}"
          become_user: "{{ web_user }}"
          environment: "{{ occ_env }}"
          # CRITICAL: Do NOT ignore errors here. This is the main failure point.

        - name: Add missing indices (optional cleanup)
          command: "{{ php_cli }} occ db:add-missing-indices"
          args:
            chdir: "{{ nextcloud_dir }}"
          become_user: "{{ web_user }}"
          environment: "{{ occ_env }}"
          ignore_errors: yes

        - name: Convert filecache bigint fields (optional cleanup)
          command: "{{ php_cli }} occ db:convert-filecache-bigint --no-interaction"
          args:
            chdir: "{{ nextcloud_dir }}"
          become_user: "{{ web_user }}"
          environment: "{{ occ_env }}"
          ignore_errors: yes

        - name: Disable maintenance mode (SUCCESS)
          command: "{{ php_cli }} occ maintenance:mode --off"
          args:
            chdir: "{{ nextcloud_dir }}"
          become_user: "{{ web_user }}"
          environment: "{{ occ_env }}"

        - name: Re-enable previously disabled 3rd party apps
          command: "{{ php_cli }} occ app:enable {{ item }}"
          args:
            chdir: "{{ nextcloud_dir }}"
          become_user: "{{ web_user }}"
          environment: "{{ occ_env }}"
          loop: "{{ disabled_apps | default([]) }}"
          when: disabled_apps is defined and disabled_apps | length > 0
          ignore_errors: yes # Don't let a single app fail the whole restart
          
        - name: Restart web service
          service:
            name: "{{ service_name }}"
            state: restarted

        - name: Check new Nextcloud version
          command: "{{ php_cli }} occ status"
          args:
            chdir: "{{ nextcloud_dir }}"
          become_user: "{{ web_user }}"
          environment: "{{ occ_env }}"
          register: occ_status_after

        - name: Display new version info
          debug:
            var: occ_status_after.stdout

        - name: ✅ Upgrade Complete
          debug:
            msg: "Nextcloud successfully upgraded!"

      # ----------------------------------------------------------------------
      # Step 4: Rescue Block (This ONLY runs if the 'block' above fails)
      # ----------------------------------------------------------------------
      rescue:
      
        - name: ❌ UPGRADE FAILED - REPORTING
          debug:
            msg: |
              !!!!!!!!!!   UPGRADE FAILED   !!!!!!!!!!
              The server is being LEFT IN MAINTENANCE MODE for manual inspection.
              
              Host: {{ inventory_hostname }}
              Please SSH to the host and check logs to fix manually.
              
              Logs to check:
              - {{ nextcloud_dir }}/data/updater.log
              - {{ nextcloud_dir }}/data/nextcloud.log
              
              Run '{{ php_cli }} occ maintenance:mode --off' manually when fixed.
              
        - name: Fail the playbook run
          fail:
            msg: "Nextcloud upgrade failed. Host is in maintenance mode."