---
# Ansible Playbook: Production Health Check
# Purpose: Comprehensive health verification for production deployment
# Usage: ansible-playbook -i inventory/production.yml playbooks/health-check.yml

- name: Production Health Check
  hosts: production_server
  become: no
  vars:
    app_url: "https://michaelschiemer.de"
    stack_name: "framework"
    health_timeout: 30
    max_retries: 10

  tasks:
    - name: Check Docker Swarm status
      shell: docker info | grep "Swarm: active"
      register: swarm_status
      failed_when: swarm_status.rc != 0
      changed_when: false

    - name: Check running services
      shell: docker service ls --filter "name={{ stack_name }}" --format "{{`{{.Name}}`}} {{`{{.Replicas}}`}}"
      register: service_list
      changed_when: false

    - name: Display service status
      debug:
        msg: "{{ service_list.stdout_lines }}"

    - name: Verify web service is running
      shell: |
        docker service ps {{ stack_name }}_web \
          --filter "desired-state=running" \
          --format "{{`{{.CurrentState}}`}}" | head -1
      register: web_state
      changed_when: false

    - name: Fail if web service not running
      fail:
        msg: "Web service is not in Running state: {{ web_state.stdout }}"
      when: "'Running' not in web_state.stdout"

    - name: Verify worker service is running
      shell: |
        docker service ps {{ stack_name }}_queue-worker \
          --filter "desired-state=running" \
          --format "{{`{{.CurrentState}}`}}" | head -1
      register: worker_state
      changed_when: false

    - name: Fail if worker service not running
      fail:
        msg: "Worker service is not in Running state: {{ worker_state.stdout }}"
      when: "'Running' not in worker_state.stdout"

    - name: Wait for application to be ready
      uri:
        url: "{{ app_url }}/health"
        validate_certs: no
        status_code: [200, 302]
        timeout: "{{ health_timeout }}"
      register: health_response
      retries: "{{ max_retries }}"
      delay: 3
      until: health_response.status in [200, 302]

    - name: Check database connectivity
      uri:
        url: "{{ app_url }}/health/database"
        validate_certs: no
        status_code: 200
        timeout: "{{ health_timeout }}"
      register: db_health
      ignore_errors: yes

    - name: Check Redis connectivity
      uri:
        url: "{{ app_url }}/health/redis"
        validate_certs: no
        status_code: 200
        timeout: "{{ health_timeout }}"
      register: redis_health
      ignore_errors: yes

    - name: Check queue system
      uri:
        url: "{{ app_url }}/health/queue"
        validate_certs: no
        status_code: 200
        timeout: "{{ health_timeout }}"
      register: queue_health
      ignore_errors: yes

    - name: Get service replicas count
      shell: |
        docker service ls --filter "name={{ stack_name }}_web" --format "{{`{{.Replicas}}`}}"
      register: replicas
      changed_when: false

    - name: Check for service errors
      shell: |
        docker service ps {{ stack_name }}_web --filter "desired-state=running" | grep -c Error || true
      register: error_count
      changed_when: false

    - name: Warn if errors detected
      debug:
        msg: "⚠️  Warning: {{ error_count.stdout }} errors detected in service logs"
      when: error_count.stdout | int > 0

    - name: Display health check summary
      debug:
        msg: |
          ✅ Health Check Summary:

          Services:
          - Web Service: {{ web_state.stdout }}
          - Worker Service: {{ worker_state.stdout }}
          - Replicas: {{ replicas.stdout }}

          Endpoints:
          - Application: {{ health_response.status }}
          - Database: {{ db_health.status | default('SKIPPED') }}
          - Redis: {{ redis_health.status | default('SKIPPED') }}
          - Queue: {{ queue_health.status | default('SKIPPED') }}

          Errors: {{ error_count.stdout }}

    - name: Overall health assessment
      debug:
        msg: "✅ All health checks PASSED"
      when:
        - health_response.status in [200, 302]
        - error_count.stdout | int == 0

    - name: Fail if critical health checks failed
      fail:
        msg: "❌ Health check FAILED - manual intervention required"
      when: health_response.status not in [200, 302]