|
| 1 | +# Arctic Text2SQL Agent - Production Environment Override |
| 2 | +# |
| 3 | +# Usage: docker-compose -f docker-compose.yml -f docker-compose.prod.yml up -d |
| 4 | +# |
| 5 | +# This file overrides the base docker-compose.yml for production deployments. |
| 6 | + |
| 7 | +services: |
| 8 | + api: |
| 9 | + image: ghcr.io/sakeeb91/arctic-text2sql-agent:${IMAGE_TAG:-latest} |
| 10 | + build: !reset null |
| 11 | + environment: |
| 12 | + # Database |
| 13 | + - DATABASE_URL=${DATABASE_URL} |
| 14 | + # HuggingFace |
| 15 | + - HUGGINGFACE_TOKEN=${HUGGINGFACE_TOKEN} |
| 16 | + - TEXT2SQL_MODEL=${TEXT2SQL_MODEL:-Snowflake/Arctic-Text2SQL-R1-7B} |
| 17 | + - MODEL_DEVICE=${MODEL_DEVICE:-cuda} |
| 18 | + # API - Production settings |
| 19 | + - API_HOST=0.0.0.0 |
| 20 | + - API_PORT=8000 |
| 21 | + - API_DEBUG=false |
| 22 | + - CORS_ORIGINS=${CORS_ORIGINS} |
| 23 | + # Agent |
| 24 | + - AGENT_MAX_STEPS=${AGENT_MAX_STEPS:-5} |
| 25 | + - AGENT_MIN_CONFIDENCE=${AGENT_MIN_CONFIDENCE:-0.8} |
| 26 | + - AGENT_VERBOSITY=0 |
| 27 | + # Logging |
| 28 | + - LOG_LEVEL=INFO |
| 29 | + - LOG_FORMAT=json |
| 30 | + # Cache |
| 31 | + - REDIS_URL=redis://redis:6379/0 |
| 32 | + # Security |
| 33 | + - SECRET_KEY=${SECRET_KEY} |
| 34 | + # Rate Limiting |
| 35 | + - RATE_LIMIT_REQUESTS=${RATE_LIMIT_REQUESTS:-100} |
| 36 | + - RATE_LIMIT_WINDOW=${RATE_LIMIT_WINDOW:-60} |
| 37 | + volumes: |
| 38 | + # Production: no source mounts, only data volumes |
| 39 | + - app-data:/app/data |
| 40 | + - huggingface-cache:/home/appuser/.cache/huggingface |
| 41 | + deploy: |
| 42 | + mode: replicated |
| 43 | + replicas: ${REPLICAS:-2} |
| 44 | + resources: |
| 45 | + limits: |
| 46 | + memory: 32G |
| 47 | + reservations: |
| 48 | + memory: 16G |
| 49 | + update_config: |
| 50 | + parallelism: 1 |
| 51 | + delay: 30s |
| 52 | + failure_action: rollback |
| 53 | + order: start-first |
| 54 | + rollback_config: |
| 55 | + parallelism: 1 |
| 56 | + delay: 10s |
| 57 | + restart_policy: |
| 58 | + condition: any |
| 59 | + delay: 5s |
| 60 | + max_attempts: 3 |
| 61 | + window: 120s |
| 62 | + labels: |
| 63 | + - "environment=production" |
| 64 | + - "version=${IMAGE_TAG:-latest}" |
| 65 | + logging: |
| 66 | + driver: json-file |
| 67 | + options: |
| 68 | + max-size: "100m" |
| 69 | + max-file: "5" |
| 70 | + |
| 71 | + db: |
| 72 | + environment: |
| 73 | + POSTGRES_USER: ${POSTGRES_USER} |
| 74 | + POSTGRES_PASSWORD: ${POSTGRES_PASSWORD} |
| 75 | + POSTGRES_DB: ${POSTGRES_DB:-text2sql_prod} |
| 76 | + deploy: |
| 77 | + resources: |
| 78 | + limits: |
| 79 | + memory: 4G |
| 80 | + reservations: |
| 81 | + memory: 2G |
| 82 | + labels: |
| 83 | + - "environment=production" |
| 84 | + logging: |
| 85 | + driver: json-file |
| 86 | + options: |
| 87 | + max-size: "50m" |
| 88 | + max-file: "3" |
| 89 | + |
| 90 | + redis: |
| 91 | + deploy: |
| 92 | + resources: |
| 93 | + limits: |
| 94 | + memory: 1G |
| 95 | + reservations: |
| 96 | + memory: 512M |
| 97 | + labels: |
| 98 | + - "environment=production" |
| 99 | + logging: |
| 100 | + driver: json-file |
| 101 | + options: |
| 102 | + max-size: "20m" |
| 103 | + max-file: "3" |
0 commit comments