runway.yaml

# Runway Platform Deployment Configuration
# Production deployment for StackMemory MCP Server

name: stackmemory-mcp
version: 1.0.0
description: "Lossless memory runtime for AI coding tools with authentication"

# Global environment variables (can be overridden per service)
environment:
  NODE_ENV: production
  LOG_LEVEL: info
  ENABLE_ANALYTICS: "true"
  ENABLE_LINEAR_SYNC: "true"
  ENABLE_WEBSOCKET: "true"

services:
  # Main MCP Server Application
  - name: mcp-server
    type: web
    runtime: node20
    dockerfile: Dockerfile.runway
    
    build:
      command: npm run build
      environment:
        NODE_ENV: production
    
    start:
      command: node --max-old-space-size=2048 dist/src/runway/index.js
      
    health_check:
      path: /health
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 60s
    
    environment:
      # Injected from Runway secrets
      DATABASE_URL: ${DATABASE_URL}
      REDIS_URL: ${REDIS_URL}
      AUTH0_DOMAIN: ${AUTH0_DOMAIN}
      AUTH0_AUDIENCE: ${AUTH0_AUDIENCE}
      AUTH0_CLIENT_ID: ${AUTH0_CLIENT_ID}
      AUTH0_CLIENT_SECRET: ${AUTH0_CLIENT_SECRET}
      JWT_SECRET: ${JWT_SECRET}
      DATADOG_API_KEY: ${DATADOG_API_KEY}
      SENTRY_DSN: ${SENTRY_DSN}
      
      # Service configuration
      PORT: 8080
      DATABASE_POOL_SIZE: 20
      RATE_LIMIT_FREE: 100
      RATE_LIMIT_PRO: 1000
      RATE_LIMIT_ENTERPRISE: 10000
      CORS_ORIGINS: "https://claude.ai,https://app.stackmemory.com"
    
    resources:
      cpu: 2000m      # 2 vCPUs
      memory: 4096Mi  # 4GB RAM
      storage: 20Gi   # 20GB disk
    
    scaling:
      min_instances: 2
      max_instances: 20
      target_cpu_utilization: 70
      target_memory_utilization: 80
      scale_down_delay: 300s
      scale_up_delay: 30s
    
    ports:
      - protocol: TCP
        port: 8080
        expose: true
        
    domains:
      - mcp.stackmemory.com
      - stackmemory-mcp.runway.app
    
    tls:
      enabled: true
      redirect_http: true
      min_version: "1.2"
      
  # Background Worker for Queue Processing
  - name: queue-worker
    type: worker
    runtime: node20
    dockerfile: Dockerfile.runway
    
    start:
      command: node dist/src/runway/queue/worker.js
    
    environment:
      DATABASE_URL: ${DATABASE_URL}
      REDIS_URL: ${REDIS_URL}
      WORKER_CONCURRENCY: 5
      WORKER_MAX_JOBS: 100
    
    resources:
      cpu: 1000m
      memory: 2048Mi
    
    scaling:
      min_instances: 1
      max_instances: 10
      target_queue_size: 100
      scale_down_delay: 600s

  # Database Migration Job
  - name: db-migrate
    type: job
    runtime: node20
    dockerfile: Dockerfile.runway
    
    command: node dist/src/runway/database/migrate.js
    
    environment:
      DATABASE_URL: ${DATABASE_URL}
    
    resources:
      cpu: 500m
      memory: 1024Mi
    
    schedule: "@deployment" # Run on each deployment
    timeout: 300s
    retries: 3

databases:
  # PostgreSQL Database
  - name: postgres
    type: postgresql
    version: "15"
    plan: production-4vcpu-8gb
    
    configuration:
      max_connections: 200
      shared_buffers: 2GB
      effective_cache_size: 6GB
      work_mem: 10MB
      maintenance_work_mem: 512MB
      random_page_cost: 1.1
      effective_io_concurrency: 200
      
    backup:
      enabled: true
      schedule: "0 2 * * *" # Daily at 2 AM
      retention: 30 # Keep 30 days of backups
      type: continuous # Point-in-time recovery
    
    high_availability:
      enabled: true
      replicas: 2
      regions:
        - us-east-1
        - us-west-2
    
    monitoring:
      enabled: true
      slow_query_log: true
      slow_query_threshold: 1000ms

caches:
  # Redis Cache
  - name: redis
    type: redis
    version: "7.0"
    plan: production-2vcpu-4gb
    
    configuration:
      maxmemory: 4GB
      maxmemory_policy: allkeys-lru
      appendonly: yes
      appendfsync: everysec
      
    high_availability:
      enabled: true
      replicas: 2
      failover: automatic
    
    persistence:
      enabled: true
      type: aof # Append-only file

storage:
  # Object Storage for file uploads
  - name: uploads
    type: s3
    size: 100Gi
    
    lifecycle:
      - rule: delete-old-temp-files
        prefix: /tmp/
        expiration_days: 7
      
      - rule: archive-old-contexts
        prefix: /contexts/
        transition_to_glacier_days: 90
    
    cors:
      - allowed_origins:
          - "https://app.stackmemory.com"
          - "https://claude.ai"
        allowed_methods:
          - GET
          - PUT
          - POST
        allowed_headers:
          - "*"
        max_age_seconds: 3600

monitoring:
  # DataDog Integration
  datadog:
    enabled: true
    api_key: ${DATADOG_API_KEY}
    tags:
      - "env:production"
      - "service:stackmemory-mcp"
    
    apm:
      enabled: true
      sample_rate: 0.1
    
    logs:
      enabled: true
      include_patterns:
        - "*.log"
        - "stdout"
        - "stderr"
    
    metrics:
      enabled: true
      custom_metrics: true
  
  # Sentry Error Tracking
  sentry:
    enabled: true
    dsn: ${SENTRY_DSN}
    environment: production
    traces_sample_rate: 0.1
    profiles_sample_rate: 0.1
    
  # CloudWatch Logs
  cloudwatch:
    enabled: true
    log_group: /runway/stackmemory-mcp
    retention_days: 30
    
alerts:
  # PagerDuty Integration
  pagerduty:
    enabled: true
    integration_key: ${PAGERDUTY_KEY}
    
  rules:
    - name: high-error-rate
      condition: "error_rate > 0.05"
      duration: 5m
      severity: critical
      
    - name: high-latency
      condition: "p95_latency > 2000"
      duration: 10m
      severity: warning
      
    - name: database-connection-pool
      condition: "db_connections_used / db_connections_total > 0.9"
      duration: 5m
      severity: warning
      
    - name: memory-usage
      condition: "memory_used / memory_total > 0.9"
      duration: 10m
      severity: warning
      
    - name: disk-usage
      condition: "disk_used / disk_total > 0.85"
      duration: 15m
      severity: warning

security:
  # Web Application Firewall
  waf:
    enabled: true
    mode: blocking
    rules:
      - owasp-top-10
      - known-bad-inputs
      - sql-injection
      - xss
    
    rate_limiting:
      enabled: true
      requests_per_minute: 1000
      burst: 100
    
    ip_whitelist:
      - "35.235.240.0/20" # Claude.ai IP range
    
    geo_blocking:
      enabled: false
      blocked_countries: []
  
  # DDoS Protection
  ddos_protection:
    enabled: true
    mode: automatic
    threshold: 10000 # requests per second
  
  # Secrets Management
  secrets:
    provider: runway-vault
    auto_rotate: true
    rotation_days: 90

deployment:
  strategy: rolling
  max_surge: 2
  max_unavailable: 1
  
  pre_deployment:
    - name: database-backup
      command: ./scripts/backup-db.sh
      
    - name: run-tests
      command: npm test
      
  post_deployment:
    - name: smoke-tests
      command: ./scripts/smoke-test.sh
      timeout: 300s
      
    - name: notify-team
      command: ./scripts/notify-deployment.sh
  
  rollback:
    automatic: true
    on_error_rate: 0.1
    on_latency_p99: 5000
    
  canary:
    enabled: true
    percentage: 10
    duration: 30m
    metrics:
      - error_rate
      - latency_p95
      - success_rate

compliance:
  # GDPR Compliance
  gdpr:
    enabled: true
    data_retention_days: 365
    right_to_be_forgotten: true
    data_portability: true
  
  # SOC2 Compliance
  soc2:
    enabled: true
    audit_logging: true
    encryption_at_rest: true
    encryption_in_transit: true

cost_optimization:
  # Auto-scaling policies to optimize costs
  quiet_hours:
    enabled: true
    schedule: "0 22 * * 1-5" # 10 PM weekdays
    min_instances: 1
    
  weekend_scaling:
    enabled: true
    schedule: "0 0 * * 6" # Saturday midnight
    min_instances: 1
    max_instances: 5

# Runway CLI commands
commands:
  deploy: runway deploy --environment production --wait
  status: runway status stackmemory-mcp
  logs: runway logs stackmemory-mcp --tail 100
  scale: runway scale stackmemory-mcp --replicas
  rollback: runway rollback stackmemory-mcp --to-previous