multitenetsaas/.github/workflows/monitoring.yml

name: System Monitoring

on:
  schedule:
    - cron: '0 */6 * * *'  # Every 6 hours
  workflow_dispatch:
  push:
    branches: [ main, develop ]

jobs:
  health-checks:
    name: Health Checks
    runs-on: ubuntu-latest

    steps:
    - name: Checkout code
      uses: actions/checkout@v4

    - name: Run production health checks
      run: |
        # API Health
        curl -f https://api.malaysian-sme-platform.com/health/ || exit 1
        curl -f https://api.malaysian-sme-platform.com/api/health/ || exit 1

        # Application Health
        curl -f https://app.malaysian-sme-platform.com/ || exit 1

        # Database Health
        curl -f https://api.malaysian-sme-platform.com/api/health/database/ || exit 1

        # Cache Health
        curl -f https://api.malaysian-sme-platform.com/api/health/cache/ || exit 1

    - name: Run staging health checks
      run: |
        curl -f https://staging.malaysian-sme-platform.com/health/ || exit 1
        curl -f https://staging.malaysian-sme-platform.com/api/health/ || exit 1

  performance-monitoring:
    name: Performance Monitoring
    runs-on: ubuntu-latest

    steps:
    - name: Checkout code
      uses: actions/checkout@v4

    - name: Set up k6
      uses: grafana/k6-action@v0.3.0
      with:
        filename: tests/performance/monitoring.js

    - name: Run performance monitoring
      run: |
        cd tests/performance
        k6 run monitoring.js \
          --env PROD_URL=https://api.malaysian-sme-platform.com \
          --env STAGING_URL=https://staging.malaysian-sme-platform.com

    - name: Upload performance results
      uses: actions/upload-artifact@v3
      with:
        name: performance-monitoring-results
        path: tests/performance/results/

  database-monitoring:
    name: Database Monitoring
    runs-on: ubuntu-latest

    steps:
    - name: Checkout code
      uses: actions/checkout@v4

    - name: Set up Python
      uses: actions/setup-python@v4
      with:
        python-version: 3.10

    - name: Install dependencies
      run: |
        python -m pip install --upgrade pip
        pip install psycopg2-binary pandas matplotlib

    - name: Run database health checks
      env:
        DATABASE_URL: ${{ secrets.PRODUCTION_DATABASE_URL }}
      run: |
        python scripts/database-health-check.py

    - name: Generate database metrics report
      env:
        DATABASE_URL: ${{ secrets.PRODUCTION_DATABASE_URL }}
      run: |
        python scripts/database-metrics.py --output database-metrics.json

    - name: Upload database reports
      uses: actions/upload-artifact@v3
      with:
        name: database-monitoring-reports
        path: |
          database-metrics.json
          database-health-report.json

  cache-monitoring:
    name: Cache Monitoring
    runs-on: ubuntu-latest

    steps:
    - name: Checkout code
      uses: actions/checkout@v4

    - name: Set up Python
      uses: actions/setup-python@v4
      with:
        python-version: 3.10

    - name: Install dependencies
      run: |
        python -m pip install --upgrade pip
        pip install redis pandas

    - name: Run Redis health checks
      env:
        REDIS_URL: ${{ secrets.PRODUCTION_REDIS_URL }}
      run: |
        python scripts/redis-health-check.py

    - name: Generate cache metrics report
      env:
        REDIS_URL: ${{ secrets.PRODUCTION_REDIS_URL }}
      run: |
        python scripts/cache-metrics.py --output cache-metrics.json

    - name: Upload cache reports
      uses: actions/upload-artifact@v3
      with:
        name: cache-monitoring-reports
        path: |
          cache-metrics.json
          redis-health-report.json

  log-monitoring:
    name: Log Monitoring
    runs-on: ubuntu-latest

    steps:
    - name: Checkout code
      uses: actions/checkout@v4

    - name: Analyze application logs
      env:
        LOGS_ACCESS_KEY: ${{ secrets.LOGS_ACCESS_KEY }}
      run: |
        python scripts/log-analysis.py \
          --hours 6 \
          --error-threshold 10 \
          --warning-threshold 50 \
          --output log-analysis-report.json

    - name: Check for critical errors
      run: |
        python scripts/critical-error-check.py \
          --hours 1 \
          --notification-webhook ${{ secrets.SLACK_WEBHOOK }}

    - name: Upload log reports
      uses: actions/upload-artifact@v3
      with:
        name: log-monitoring-reports
        path: |
          log-analysis-report.json
          error-summary.json

  resource-monitoring:
    name: Resource Monitoring
    runs-on: ubuntu-latest

    steps:
    - name: Checkout code
      uses: actions/checkout@v4

    - name: Monitor system resources
      env:
        MONITORING_API_KEY: ${{ secrets.MONITORING_API_KEY }}
      run: |
        python scripts/resource-monitoring.py \
          --output resource-metrics.json

    - name: Check resource thresholds
      run: |
        python scripts/resource-threshold-check.py \
          --cpu-threshold 80 \
          --memory-threshold 85 \
          --disk-threshold 90 \
          --output threshold-report.json

    - name: Upload resource reports
      uses: actions/upload-artifact@v3
      with:
        name: resource-monitoring-reports
        path: |
          resource-metrics.json
          threshold-report.json

  uptime-monitoring:
    name: Uptime Monitoring
    runs-on: ubuntu-latest

    steps:
    - name: Checkout code
      uses: actions/checkout@v4

    - name: Check service uptime
      run: |
        python scripts/uptime-check.py \
          --services api,app,admin,static \
          --timeout 30 \
          --output uptime-report.json

    - name: Verify SSL certificates
      run: |
        python scripts/ssl-check.py \
          --domains api.malaysian-sme-platform.com,app.malaysian-sme-platform.com \
          --output ssl-report.json

    - name: Upload uptime reports
      uses: actions/upload-artifact@v3
      with:
        name: uptime-monitoring-reports
        path: |
          uptime-report.json
          ssl-report.json

  backup-monitoring:
    name: Backup Monitoring
    runs-on: ubuntu-latest

    steps:
    - name: Checkout code
      uses: actions/checkout@v4

    - name: Check backup status
      env:
        BACKUP_ACCESS_KEY: ${{ secrets.BACKUP_ACCESS_KEY }}
      run: |
        python scripts/backup-check.py \
          --hours 24 \
          --output backup-report.json

    - name: Verify backup integrity
      run: |
        python scripts/backup-integrity.py \
          --verify-latest 3 \
          --output integrity-report.json

    - name: Upload backup reports
      uses: actions/upload-artifact@v3
      with:
        name: backup-monitoring-reports
        path: |
          backup-report.json
          integrity-report.json

  security-monitoring:
    name: Security Monitoring
    runs-on: ubuntu-latest

    steps:
    - name: Checkout code
      uses: actions/checkout@v4

    - name: Check for security events
      env:
        SECURITY_API_KEY: ${{ secrets.SECURITY_API_KEY }}
      run: |
        python scripts/security-monitoring.py \
          --hours 6 \
          --output security-events.json

    - name: Analyze authentication patterns
      env:
        AUTH_LOGS_ACCESS_KEY: ${{ secrets.AUTH_LOGS_ACCESS_KEY }}
      run: |
        python scripts/auth-pattern-analysis.py \
          --hours 24 \
          --output auth-patterns.json

    - name: Upload security reports
      uses: actions/upload-artifact@v3
      with:
        name: security-monitoring-reports
        path: |
          security-events.json
          auth-patterns.json

  monitoring-dashboard:
    name: Monitoring Dashboard
    runs-on: ubuntu-latest
    needs: [health-checks, performance-monitoring, database-monitoring, cache-monitoring, log-monitoring, resource-monitoring, uptime-monitoring, backup-monitoring, security-monitoring]
    if: always()

    steps:
    - name: Download all reports
      uses: actions/download-artifact@v3

    - name: Generate monitoring dashboard
      run: |
        python scripts/generate-monitoring-dashboard.py

    - name: Upload monitoring dashboard
      uses: actions/upload-artifact@v3
      with:
        name: monitoring-dashboard
        path: monitoring-dashboard.html

    - name: Send monitoring summary to Slack
      uses: 8398a7/action-slack@v3
      with:
        status: ${{ job.status }}
        channel: '#monitoring'
        webhook_url: ${{ secrets.SLACK_WEBHOOK }}
      env:
        SLACK_WEBHOOK_URL: ${{ secrets.SLACK_WEBHOOK }}

  alerting:
    name: Alerting
    runs-on: ubuntu-latest
    needs: monitoring-dashboard
    if: failure()

    steps:
    - name: Create alert issue
      uses: actions/github-script@v6
      with:
        script: |
          github.rest.issues.create({
            owner: context.repo.owner,
            repo: context.repo.repo,
            title: `🚨 Monitoring Alert - ${{ github.run_number }}`,
            body: `Monitoring checks failed for run #${{ github.run_number }}.

          **Time:** ${{ github.event_name }} at ${{ github.run_started_at }}
          **Repository:** ${{ github.repository }}

          Please review the monitoring reports and investigate the issues.

          📋 **Monitoring Reports:**
          - [Health Checks](https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }})
          - [Performance Monitoring](https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }})
          - [Database Monitoring](https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }})
          - [Cache Monitoring](https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }})
          - [Log Monitoring](https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }})
          - [Resource Monitoring](https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }})
          - [Uptime Monitoring](https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }})
          - [Backup Monitoring](https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }})
          - [Security Monitoring](https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }})

          🎯 **Immediate Actions:**
          1. Review failed monitoring checks
          2. Investigate service availability
          3. Check system resources
          4. Verify backup integrity
          5. Address any security events

          This issue was automatically created by the monitoring system.`,
            labels: ['monitoring', 'alert', 'priority-critical']
          });

    - name: Send emergency notification
      uses: 8398a7/action-slack@v3
      with:
        status: failure
        channel: '#emergency'
        webhook_url: ${{ secrets.EMERGENCY_SLACK_WEBHOOK }}
      env:
        SLACK_WEBHOOK_URL: ${{ secrets.EMERGENCY_SLACK_WEBHOOK }}