---
# Diagnose Gitea Timeout - Live während Request
# Führt alle Checks während eines tatsächlichen Requests durch
- name: Diagnose Gitea Timeout During Request
  hosts: production
  gather_facts: yes
  become: no
  vars:
    gitea_stack_path: "{{ stacks_base_path }}/gitea"
    traefik_stack_path: "{{ stacks_base_path }}/traefik"
    gitea_url: "https://{{ gitea_domain }}"
    test_duration_seconds: 60  # Wie lange wir testen
    test_timestamp: "{{ ansible_date_time.epoch }}"

  tasks:
    - name: Display diagnostic plan
      ansible.builtin.debug:
        msg: |
          ================================================================================
          GITEA TIMEOUT DIAGNOSE - LIVE WÄHREND REQUEST
          ================================================================================
          
          Diese Diagnose führt alle Checks während eines tatsächlichen Requests durch:
          
          1. Docker Stats (CPU/RAM/IO) während Request
          2. Gitea Logs (DB-Timeouts, Panics, "context deadline exceeded")
          3. Postgres Logs (Connection issues)
          4. Traefik Logs ("backend connection error", "EOF")
          5. Direkter Test Traefik → Gitea
          
          Test-Dauer: {{ test_duration_seconds }} Sekunden
          Timestamp: {{ test_timestamp }}
          ================================================================================

    - name: Get initial container stats (baseline)
      ansible.builtin.shell: |
        docker stats --no-stream --format "table {{ '{{' }}.Name{{ '}}' }}\t{{ '{{' }}.CPUPerc{{ '}}' }}\t{{ '{{' }}.MemUsage{{ '}}' }}\t{{ '{{' }}.NetIO{{ '}}' }}\t{{ '{{' }}.BlockIO{{ '}}' }}" gitea gitea-postgres gitea-redis traefik 2>/dev/null || echo "Stats collection failed"
      register: initial_stats
      changed_when: false

    - name: Start collecting Docker stats in background
      ansible.builtin.shell: |
        timeout {{ test_duration_seconds }} docker stats --format "{{ '{{' }}.Name{{ '}}' }},{{ '{{' }}.CPUPerc{{ '}}' }},{{ '{{' }}.MemUsage{{ '}}' }},{{ '{{' }}.NetIO{{ '}}' }},{{ '{{' }}.BlockIO{{ '}}' }}" gitea gitea-postgres gitea-redis traefik 2>/dev/null | while read line; do
          echo "[$(date '+%Y-%m-%d %H:%M:%S.%3N')] $line"
        done > /tmp/gitea_stats_{{ test_timestamp }}.log 2>&1 &
        STATS_PID=$!
        echo $STATS_PID
      register: stats_pid
      changed_when: false

    - name: Start collecting Gitea logs in background
      ansible.builtin.shell: |
        cd {{ gitea_stack_path }}
        timeout {{ test_duration_seconds }} docker compose logs -f gitea 2>&1 | while read line; do
          echo "[$(date '+%Y-%m-%d %H:%M:%S.%3N')] $line"
        done > /tmp/gitea_logs_{{ test_timestamp }}.log 2>&1 &
        echo $!
      register: gitea_logs_pid
      changed_when: false

    - name: Start collecting Postgres logs in background
      ansible.builtin.shell: |
        cd {{ gitea_stack_path }}
        timeout {{ test_duration_seconds }} docker compose logs -f gitea-postgres 2>&1 | while read line; do
          echo "[$(date '+%Y-%m-%d %H:%M:%S.%3N')] $line"
        done > /tmp/postgres_logs_{{ test_timestamp }}.log 2>&1 &
        echo $!
      register: postgres_logs_pid
      changed_when: false

    - name: Start collecting Traefik logs in background
      ansible.builtin.shell: |
        cd {{ traefik_stack_path }}
        timeout {{ test_duration_seconds }} docker compose logs -f traefik 2>&1 | while read line; do
          echo "[$(date '+%Y-%m-%d %H:%M:%S.%3N')] $line"
        done > /tmp/traefik_logs_{{ test_timestamp }}.log 2>&1 &
        echo $!
      register: traefik_logs_pid
      changed_when: false

    - name: Wait a moment for log collection to start
      ansible.builtin.pause:
        seconds: 2

    - name: Trigger Gitea request via Traefik (with timeout)
      ansible.builtin.shell: |
        echo "[$(date '+%Y-%m-%d %H:%M:%S.%3N')] Starting request to {{ gitea_url }}/api/healthz"
        timeout 35 curl -k -v -s -o /tmp/gitea_response_{{ test_timestamp }}.log -w "\nHTTP_CODE:%{http_code}\nTIME_TOTAL:%{time_total}\nTIME_CONNECT:%{time_connect}\nTIME_STARTTRANSFER:%{time_starttransfer}\n" "{{ gitea_url }}/api/healthz" 2>&1 | tee /tmp/gitea_curl_{{ test_timestamp }}.log
        echo "[$(date '+%Y-%m-%d %H:%M:%S.%3N')] Request completed"
      register: gitea_request
      changed_when: false
      failed_when: false

    - name: Test direct connection Traefik → Gitea (parallel)
      ansible.builtin.shell: |
        echo "[$(date '+%Y-%m-%d %H:%M:%S.%3N')] Starting direct test Traefik → Gitea"
        cd {{ traefik_stack_path }}
        timeout 35 docker compose exec -T traefik wget -qO- --timeout=30 http://gitea:3000/api/healthz 2>&1 | tee /tmp/traefik_gitea_direct_{{ test_timestamp }}.log || echo "DIRECT_TEST_FAILED" > /tmp/traefik_gitea_direct_{{ test_timestamp }}.log
        echo "[$(date '+%Y-%m-%d %H:%M:%S.%3N')] Direct test completed"
      register: traefik_direct_test
      changed_when: false
      failed_when: false

    - name: Wait for log collection to complete
      ansible.builtin.pause:
        seconds: "{{ test_duration_seconds - 5 }}"

    - name: Stop background processes
      ansible.builtin.shell: |
        pkill -f "docker.*stats.*gitea" || true
        pkill -f "docker compose logs.*gitea" || true
        pkill -f "docker compose logs.*postgres" || true
        pkill -f "docker compose logs.*traefik" || true
        sleep 2
      changed_when: false
      failed_when: false

    - name: Collect stats results
      ansible.builtin.slurp:
        src: "/tmp/gitea_stats_{{ test_timestamp }}.log"
      register: stats_results
      changed_when: false
      failed_when: false

    - name: Collect Gitea logs results
      ansible.builtin.slurp:
        src: "/tmp/gitea_logs_{{ test_timestamp }}.log"
      register: gitea_logs_results
      changed_when: false
      failed_when: false

    - name: Collect Postgres logs results
      ansible.builtin.slurp:
        src: "/tmp/postgres_logs_{{ test_timestamp }}.log"
      register: postgres_logs_results
      changed_when: false
      failed_when: false

    - name: Collect Traefik logs results
      ansible.builtin.slurp:
        src: "/tmp/traefik_logs_{{ test_timestamp }}.log"
      register: traefik_logs_results
      changed_when: false
      failed_when: false

    - name: Get request result
      ansible.builtin.slurp:
        src: "/tmp/gitea_curl_{{ test_timestamp }}.log"
      register: request_result
      changed_when: false
      failed_when: false

    - name: Get direct test result
      ansible.builtin.slurp:
        src: "/tmp/traefik_gitea_direct_{{ test_timestamp }}.log"
      register: direct_test_result
      changed_when: false
      failed_when: false

    - name: Analyze stats for high CPU/Memory/IO
      ansible.builtin.shell: |
        if [ -f /tmp/gitea_stats_{{ test_timestamp }}.log ]; then
          echo "=== STATS SUMMARY ==="
          echo "Total samples: $(wc -l < /tmp/gitea_stats_{{ test_timestamp }}.log)"
          echo ""
          echo "=== HIGH CPU (>80%) ==="
          grep -E "gitea|gitea-postgres" /tmp/gitea_stats_{{ test_timestamp }}.log | awk -F',' '{cpu=$2; gsub(/%/, "", cpu); if (cpu+0 > 80) print $0}' | head -10 || echo "No high CPU usage found"
          echo ""
          echo "=== MEMORY USAGE ==="
          grep -E "gitea" /tmp/gitea_stats_{{ test_timestamp }}.log | tail -5 || echo "No memory stats"
          echo ""
          echo "=== NETWORK IO ==="
          grep -E "gitea" /tmp/gitea_stats_{{ test_timestamp }}.log | tail -5 || echo "No network activity"
        else
          echo "Stats file not found"
        fi
      register: stats_analysis
      changed_when: false
      failed_when: false

    - name: Analyze Gitea logs for errors
      ansible.builtin.shell: |
        if [ -f /tmp/gitea_logs_{{ test_timestamp }}.log ]; then
          echo "=== DB-TIMEOUTS / CONNECTION ERRORS ==="
          grep -iE "timeout|deadline exceeded|connection.*failed|database.*error|postgres.*error|context.*deadline" /tmp/gitea_logs_{{ test_timestamp }}.log | tail -20 || echo "No DB-timeouts found"
          echo ""
          echo "=== PANICS / FATAL ERRORS ==="
          grep -iE "panic|fatal|error.*fatal" /tmp/gitea_logs_{{ test_timestamp }}.log | tail -10 || echo "No panics found"
          echo ""
          echo "=== SLOW QUERIES / PERFORMANCE ==="
          grep -iE "slow|performance|took.*ms|duration" /tmp/gitea_logs_{{ test_timestamp }}.log | tail -10 || echo "No slow queries found"
          echo ""
          echo "=== RECENT LOG ENTRIES (last 10) ==="
          tail -10 /tmp/gitea_logs_{{ test_timestamp }}.log || echo "No recent logs"
        else
          echo "Gitea logs file not found"
        fi
      register: gitea_logs_analysis
      changed_when: false
      failed_when: false

    - name: Analyze Postgres logs for errors
      ansible.builtin.shell: |
        if [ -f /tmp/postgres_logs_{{ test_timestamp }}.log ]; then
          echo "=== POSTGRES ERRORS ==="
          grep -iE "error|timeout|deadlock|connection.*refused|too many connections" /tmp/postgres_logs_{{ test_timestamp }}.log | tail -20 || echo "No Postgres errors found"
          echo ""
          echo "=== SLOW QUERIES ==="
          grep -iE "slow|duration|statement.*took" /tmp/postgres_logs_{{ test_timestamp }}.log | tail -10 || echo "No slow queries found"
          echo ""
          echo "=== RECENT LOG ENTRIES (last 10) ==="
          tail -10 /tmp/postgres_logs_{{ test_timestamp }}.log || echo "No recent logs"
        else
          echo "Postgres logs file not found"
        fi
      register: postgres_logs_analysis
      changed_when: false
      failed_when: false

    - name: Analyze Traefik logs for backend errors
      ansible.builtin.shell: |
        if [ -f /tmp/traefik_logs_{{ test_timestamp }}.log ]; then
          echo "=== BACKEND CONNECTION ERRORS ==="
          grep -iE "backend.*error|connection.*error|EOF|gitea.*error|git\.michaelschiemer\.de.*error" /tmp/traefik_logs_{{ test_timestamp }}.log | tail -20 || echo "No backend errors found"
          echo ""
          echo "=== TIMEOUT ERRORS ==="
          grep -iE "timeout|504|gateway.*timeout" /tmp/traefik_logs_{{ test_timestamp }}.log | tail -10 || echo "No timeout errors found"
          echo ""
          echo "=== RECENT LOG ENTRIES (last 10) ==="
          tail -10 /tmp/traefik_logs_{{ test_timestamp }}.log || echo "No recent logs"
        else
          echo "Traefik logs file not found"
        fi
      register: traefik_logs_analysis
      changed_when: false
      failed_when: false

    - name: Display comprehensive diagnosis
      ansible.builtin.debug:
        msg: |
          ================================================================================
          GITEA TIMEOUT DIAGNOSE - ERGEBNISSE
          ================================================================================
          
          BASELINE STATS (vor Request):
          {{ initial_stats.stdout }}
          
          REQUEST ERGEBNIS:
          {% if request_result.content is defined and request_result.content != '' %}
          {{ request_result.content | b64decode }}
          {% else %}
          Request-Ergebnis nicht verfügbar
          {% endif %}
          
          DIREKTER TEST TRAEFIK → GITEA:
          {% if direct_test_result.content is defined and direct_test_result.content != '' %}
          {{ direct_test_result.content | b64decode }}
          {% else %}
          Direkter Test-Ergebnis nicht verfügbar
          {% endif %}
          
          ================================================================================
          STATS-ANALYSE (während Request):
          ================================================================================
          {{ stats_analysis.stdout }}
          
          ================================================================================
          GITEA LOGS-ANALYSE:
          ================================================================================
          {{ gitea_logs_analysis.stdout }}
          
          ================================================================================
          POSTGRES LOGS-ANALYSE:
          ================================================================================
          {{ postgres_logs_analysis.stdout }}
          
          ================================================================================
          TRAEFIK LOGS-ANALYSE:
          ================================================================================
          {{ traefik_logs_analysis.stdout }}
          
          ================================================================================
          INTERPRETATION:
          ================================================================================
          
          {% set request_content = request_result.content | default('') | b64decode | default('') %}
          {% set direct_content = direct_test_result.content | default('') | b64decode | default('') %}
          {% set traefik_errors = traefik_logs_analysis.stdout | default('') %}
          {% set gitea_errors = gitea_logs_analysis.stdout | default('') %}
          {% set postgres_errors = postgres_logs_analysis.stdout | default('') %}
          {% set stats_content = stats_analysis.stdout | default('') %}
          
          {% if 'timeout' in request_content or '504' in request_content or 'HTTP_CODE:504' in request_content %}
          ⚠️  REQUEST HAT TIMEOUT/504:
          
          {% if 'EOF' in traefik_errors or 'backend' in traefik_errors | lower or 'connection.*error' in traefik_errors | lower %}
          → Traefik meldet Backend-Connection-Error
          → Gitea antwortet nicht auf Traefik's Verbindungsversuche
          {% endif %}
          
          {% if 'timeout' in gitea_errors | lower or 'deadline exceeded' in gitea_errors | lower %}
          → Gitea hat DB-Timeouts oder Context-Deadline-Exceeded
          → Postgres könnte blockieren oder zu langsam sein
          {% endif %}
          
          {% if 'too many connections' in postgres_errors | lower %}
          → Postgres hat zu viele Verbindungen
          → Connection Pool könnte überlastet sein
          {% endif %}
          
          {% if 'HIGH CPU' in stats_content or '>80' in stats_content %}
          → Gitea oder Postgres haben hohe CPU-Last
          → Performance-Problem, nicht Timeout-Konfiguration
          {% endif %}
          
          {% if 'DIRECT_TEST_FAILED' in direct_content or direct_content == '' %}
          → Direkter Test Traefik → Gitea schlägt fehl
          → Problem liegt bei Gitea selbst, nicht bei Traefik-Routing
          {% endif %}
          {% else %}
          ✅ REQUEST WAR ERFOLGREICH:
          → Problem tritt nur intermittierend auf
          → Prüfe Logs auf sporadische Fehler
          {% endif %}
          
          ================================================================================
          NÄCHSTE SCHRITTE:
          ================================================================================
          
          1. Prüfe ob hohe CPU/Memory bei Gitea oder Postgres
          2. Prüfe ob DB-Timeouts in Gitea-Logs
          3. Prüfe ob Postgres "too many connections" meldet
          4. Prüfe ob Traefik "backend connection error" oder "EOF" meldet
          5. Prüfe ob direkter Test Traefik → Gitea funktioniert
          
          ================================================================================

    - name: Cleanup temporary files
      ansible.builtin.file:
        path: "/tmp/gitea_{{ test_timestamp }}.log"
        state: absent
      failed_when: false