multi-instance.yaml

# DEVELOPMENT ONLY — do not deploy to production. Sets MTLS_REQUIRED=false and admin cors_origin=*.
#
# Aether Gateway Multi-Instance Testing
# Docker Compose configuration for testing horizontal scaling with session affinity
#
# This setup includes:
# - 3 gateway instances with unique IDs
# - nginx load balancer with session affinity
# - Redis (valkey) for session registry and KV store
# - RabbitMQ with Streams plugin for messaging
# - PostgreSQL for task persistence
#
# Required environment variables:
#   RABBITMQ_PASS  — RabbitMQ password (no default; must be set explicitly)
#
# Optional environment variables:
#   RABBITMQ_USER  — RabbitMQ username (default: aether)
#
# Usage:
#   RABBITMQ_PASS=yourpassword docker-compose -f server/deployments/docker-compose/multi-instance.yaml up
#   docker-compose -f server/deployments/docker-compose/multi-instance.yaml down

version: '3.8'

services:
  # =============================================================================
  # Infrastructure Services
  # =============================================================================

  # PostgreSQL - Task persistence and metadata storage
  postgres:
    image: postgres:15-alpine
    container_name: aether-postgres
    environment:
      POSTGRES_DB: aether
      POSTGRES_USER: aether
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:?POSTGRES_PASSWORD required}
    ports:
      - "5432:5432"
    volumes:
      - postgres-data:/var/lib/postgresql/data
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U aether"]
      interval: 10s
      timeout: 5s
      retries: 5
    networks:
      - aether

  # Redis (Valkey) - Session registry, KV store, checkpoint store
  redis:
    image: valkey/valkey:latest
    container_name: aether-redis
    command:
      - valkey-server
      - --bind
      - "0.0.0.0"
      - --protected-mode
      - "no"
      - --appendonly
      - "no"
      - --save
      - ""
    ports:
      - "6379:6379"
    healthcheck:
      test: ["CMD", "valkey-cli", "ping"]
      interval: 10s
      timeout: 5s
      retries: 5
    networks:
      - aether

  # RabbitMQ with Streams - Message routing backbone
  rabbitmq:
    image: ghcr.io/scitrera/rabbitmq-stream:4-management
    container_name: aether-rabbitmq
    environment:
      RABBITMQ_DEFAULT_USER: ${RABBITMQ_USER:-aether}
      RABBITMQ_DEFAULT_PASS: ${RABBITMQ_PASS:?RABBITMQ_PASS required}
    ports:
      - "5552:5552"   # Stream protocol
      - "5672:5672"   # AMQP
      - "15672:15672" # Management UI
    volumes:
      - rabbitmq-data:/var/lib/rabbitmq
      - ./nginx/rabbitmq.conf:/etc/rabbitmq/rabbitmq.conf:ro
    healthcheck:
      test: ["CMD", "rabbitmq-diagnostics", "ping"]
      interval: 30s
      timeout: 10s
      retries: 5
    networks:
      - aether

  # =============================================================================
  # Gateway Instances (3 replicas for high availability)
  # =============================================================================

  gateway-1:
    image: scitrera/aether-gateway:latest
    container_name: aether-gateway-1
    build:
      context: ../..
      dockerfile: server/Dockerfile
    environment:
      # Gateway identity
      AETHER_GATEWAY_ID: gateway-1
      GATEWAY_PORT: "50051"
      GATEWAY_HTTP_PORT: "8080"

      # PostgreSQL
      POSTGRES_HOST: postgres
      POSTGRES_PORT: "5432"
      POSTGRES_DATABASE: aether
      POSTGRES_USER: aether
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:?POSTGRES_PASSWORD required}
      POSTGRES_SSL_MODE: disable
      POSTGRES_MAX_CONNECTIONS: "25"
      POSTGRES_MAX_IDLE_CONNECTIONS: "5"

      # Redis
      REDIS_CLUSTER_ADDR_0: redis:6379
      REDIS_PASSWORD: ""
      REDIS_DB: "0"

      # RabbitMQ
      RABBITMQ_STREAM_URL: rabbitmq-stream://${RABBITMQ_USER:-aether}:${RABBITMQ_PASS:?RABBITMQ_PASS required}@rabbitmq:5552
      RABBITMQ_AMQP_URL: amqp://${RABBITMQ_USER:-aether}:${RABBITMQ_PASS:?RABBITMQ_PASS required}@rabbitmq:5672/

      # mTLS (disabled for local testing)
      MTLS_REQUIRED: "false"
      MTLS_MODE: relaxed

      # Observability
      OBSERVABILITY_LOG_LEVEL: info
      OBSERVABILITY_PROMETHEUS_PORT: "9090"

      # Cleanup settings
      CLEANUP_TASK_PURGE_INTERVAL: 24h
      CLEANUP_COMPLETED_TASK_RETENTION: 168h
      CLEANUP_FAILED_TASK_RETENTION: 336h
      CLEANUP_CANCELLED_TASK_RETENTION: 168h
      CLEANUP_RECONCILIATION_INTERVAL: 1m

      # Checkpoint settings
      CHECKPOINT_DEFAULT_TTL: 3600s
    ports:
      - "50051:50051"
      - "8081:8080"
      - "9091:9090"
    depends_on:
      postgres:
        condition: service_healthy
      redis:
        condition: service_healthy
      rabbitmq:
        condition: service_healthy
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8080/health/live"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 30s
    networks:
      - aether

  gateway-2:
    image: scitrera/aether-gateway:latest
    container_name: aether-gateway-2
    build:
      context: ../..
      dockerfile: server/Dockerfile
    environment:
      AETHER_GATEWAY_ID: gateway-2
      GATEWAY_PORT: "50051"
      GATEWAY_HTTP_PORT: "8080"
      POSTGRES_HOST: postgres
      POSTGRES_PORT: "5432"
      POSTGRES_DATABASE: aether
      POSTGRES_USER: aether
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:?POSTGRES_PASSWORD required}
      POSTGRES_SSL_MODE: disable
      POSTGRES_MAX_CONNECTIONS: "25"
      POSTGRES_MAX_IDLE_CONNECTIONS: "5"
      REDIS_CLUSTER_ADDR_0: redis:6379
      REDIS_PASSWORD: ""
      REDIS_DB: "0"
      RABBITMQ_STREAM_URL: rabbitmq-stream://${RABBITMQ_USER:-aether}:${RABBITMQ_PASS:?RABBITMQ_PASS required}@rabbitmq:5552
      RABBITMQ_AMQP_URL: amqp://${RABBITMQ_USER:-aether}:${RABBITMQ_PASS:?RABBITMQ_PASS required}@rabbitmq:5672/
      MTLS_REQUIRED: "false"
      MTLS_MODE: relaxed
      OBSERVABILITY_LOG_LEVEL: info
      OBSERVABILITY_PROMETHEUS_PORT: "9090"
      CLEANUP_TASK_PURGE_INTERVAL: 24h
      CLEANUP_COMPLETED_TASK_RETENTION: 168h
      CLEANUP_FAILED_TASK_RETENTION: 336h
      CLEANUP_CANCELLED_TASK_RETENTION: 168h
      CLEANUP_RECONCILIATION_INTERVAL: 1m
      CHECKPOINT_DEFAULT_TTL: 3600s
    ports:
      - "50052:50051"
      - "8082:8080"
      - "9092:9090"
    depends_on:
      postgres:
        condition: service_healthy
      redis:
        condition: service_healthy
      rabbitmq:
        condition: service_healthy
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8080/health/live"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 30s
    networks:
      - aether

  gateway-3:
    image: scitrera/aether-gateway:latest
    container_name: aether-gateway-3
    build:
      context: ../..
      dockerfile: server/Dockerfile
    environment:
      AETHER_GATEWAY_ID: gateway-3
      GATEWAY_PORT: "50051"
      GATEWAY_HTTP_PORT: "8080"
      POSTGRES_HOST: postgres
      POSTGRES_PORT: "5432"
      POSTGRES_DATABASE: aether
      POSTGRES_USER: aether
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:?POSTGRES_PASSWORD required}
      POSTGRES_SSL_MODE: disable
      POSTGRES_MAX_CONNECTIONS: "25"
      POSTGRES_MAX_IDLE_CONNECTIONS: "5"
      REDIS_CLUSTER_ADDR_0: redis:6379
      REDIS_PASSWORD: ""
      REDIS_DB: "0"
      RABBITMQ_STREAM_URL: rabbitmq-stream://${RABBITMQ_USER:-aether}:${RABBITMQ_PASS:?RABBITMQ_PASS required}@rabbitmq:5552
      RABBITMQ_AMQP_URL: amqp://${RABBITMQ_USER:-aether}:${RABBITMQ_PASS:?RABBITMQ_PASS required}@rabbitmq:5672/
      MTLS_REQUIRED: "false"
      MTLS_MODE: relaxed
      OBSERVABILITY_LOG_LEVEL: info
      OBSERVABILITY_PROMETHEUS_PORT: "9090"
      CLEANUP_TASK_PURGE_INTERVAL: 24h
      CLEANUP_COMPLETED_TASK_RETENTION: 168h
      CLEANUP_FAILED_TASK_RETENTION: 336h
      CLEANUP_CANCELLED_TASK_RETENTION: 168h
      CLEANUP_RECONCILIATION_INTERVAL: 1m
      CHECKPOINT_DEFAULT_TTL: 3600s
    ports:
      - "50053:50051"
      - "8083:8080"
      - "9093:9090"
    depends_on:
      postgres:
        condition: service_healthy
      redis:
        condition: service_healthy
      rabbitmq:
        condition: service_healthy
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8080/health/live"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 30s
    networks:
      - aether

  # =============================================================================
  # Load Balancer (nginx with session affinity)
  # =============================================================================

  nginx:
    image: nginx:alpine
    container_name: aether-nginx
    volumes:
      - ./nginx/nginx.conf:/etc/nginx/nginx.conf:ro
    ports:
      - "50050:50050"  # gRPC load balanced endpoint
      - "8080:8080"    # HTTP health check endpoint
    depends_on:
      gateway-1:
        condition: service_healthy
      gateway-2:
        condition: service_healthy
      gateway-3:
        condition: service_healthy
    healthcheck:
      test: ["CMD", "wget", "--quiet", "--tries=1", "--spider", "http://localhost:8080/health"]
      interval: 10s
      timeout: 5s
      retries: 5
    networks:
      - aether

# =============================================================================
# Volumes
# =============================================================================

volumes:
  postgres-data:
    driver: local
  rabbitmq-data:
    driver: local

# =============================================================================
# Networks
# =============================================================================

networks:
  aether:
    driver: bridge