synor/docker-compose.compute.yml

# Synor Compute Layer - Docker Compose
# Heterogeneous compute orchestration for AI/ML workloads

version: '3.9'

services:
  # Compute Orchestrator (schedules tasks across workers)
  compute-orchestrator:
    build:
      context: .
      dockerfile: docker/compute-node/Dockerfile
    container_name: synor-compute-orchestrator
    hostname: compute-orchestrator
    restart: unless-stopped
    environment:
      - RUST_LOG=info
      - NODE_TYPE=orchestrator
      - LISTEN_ADDR=0.0.0.0:17200
      - WORKER_PORTS=17210,17211,17212,17213
      - BALANCING_STRATEGY=balanced
      - MAX_QUEUE_SIZE=10000
    ports:
      - "17250:17200"   # Compute API
      - "17252:17202"   # Metrics/Health
    networks:
      - synor-compute-net
    volumes:
      - compute-orchestrator-data:/data/compute
    depends_on:
      - compute-worker-cpu-1
      - compute-worker-cpu-2
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:17202/health"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 30s

  # CPU Worker Node 1 (x86-64 AVX2)
  compute-worker-cpu-1:
    build:
      context: .
      dockerfile: docker/compute-node/Dockerfile
    container_name: synor-compute-worker-cpu-1
    hostname: compute-worker-cpu-1
    restart: unless-stopped
    environment:
      - RUST_LOG=info
      - NODE_TYPE=worker
      - PROCESSOR_TYPE=cpu
      - CPU_VARIANT=x86_64_avx2
      - LISTEN_ADDR=0.0.0.0:17210
      - ORCHESTRATOR_URL=http://compute-orchestrator:17200
      - MAX_CONCURRENT_TASKS=8
      - WORK_STEAL_ENABLED=true
    ports:
      - "17260:17210"   # Worker API
    networks:
      - synor-compute-net
    volumes:
      - compute-worker-cpu-1-data:/data/compute
    deploy:
      resources:
        limits:
          cpus: '4'
          memory: 8G
        reservations:
          cpus: '2'
          memory: 4G
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:17210/health"]
      interval: 30s
      timeout: 5s
      retries: 3

  # CPU Worker Node 2 (x86-64 AVX2)
  compute-worker-cpu-2:
    build:
      context: .
      dockerfile: docker/compute-node/Dockerfile
    container_name: synor-compute-worker-cpu-2
    hostname: compute-worker-cpu-2
    restart: unless-stopped
    environment:
      - RUST_LOG=info
      - NODE_TYPE=worker
      - PROCESSOR_TYPE=cpu
      - CPU_VARIANT=x86_64_avx2
      - LISTEN_ADDR=0.0.0.0:17211
      - ORCHESTRATOR_URL=http://compute-orchestrator:17200
      - MAX_CONCURRENT_TASKS=8
      - WORK_STEAL_ENABLED=true
    ports:
      - "17261:17211"   # Worker API
    networks:
      - synor-compute-net
    volumes:
      - compute-worker-cpu-2-data:/data/compute
    deploy:
      resources:
        limits:
          cpus: '4'
          memory: 8G
        reservations:
          cpus: '2'
          memory: 4G
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:17211/health"]
      interval: 30s
      timeout: 5s
      retries: 3

  # WASM Worker (browser-compatible compute)
  compute-worker-wasm:
    build:
      context: .
      dockerfile: docker/compute-node/Dockerfile
    container_name: synor-compute-worker-wasm
    hostname: compute-worker-wasm
    restart: unless-stopped
    environment:
      - RUST_LOG=info
      - NODE_TYPE=worker
      - PROCESSOR_TYPE=wasm
      - LISTEN_ADDR=0.0.0.0:17212
      - ORCHESTRATOR_URL=http://compute-orchestrator:17200
      - MAX_CONCURRENT_TASKS=4
    ports:
      - "17262:17212"   # Worker API
    networks:
      - synor-compute-net
    volumes:
      - compute-worker-wasm-data:/data/compute
    deploy:
      resources:
        limits:
          cpus: '2'
          memory: 2G
        reservations:
          cpus: '1'
          memory: 1G
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:17212/health"]
      interval: 30s
      timeout: 5s
      retries: 3

  # Spot Market Service
  compute-spot-market:
    build:
      context: .
      dockerfile: docker/compute-node/Dockerfile
    container_name: synor-compute-spot-market
    hostname: compute-spot-market
    restart: unless-stopped
    environment:
      - RUST_LOG=info
      - NODE_TYPE=market
      - LISTEN_ADDR=0.0.0.0:17220
      - ORCHESTRATOR_URL=http://compute-orchestrator:17200
      - AUCTION_INTERVAL_MS=5000
      - MIN_BID_MICRO=100
    ports:
      - "17270:17220"   # Market API
    networks:
      - synor-compute-net
    volumes:
      - compute-spot-market-data:/data/compute
    depends_on:
      - compute-orchestrator
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:17220/health"]
      interval: 30s
      timeout: 5s
      retries: 3

  # Redis for task queue and caching
  compute-redis:
    image: redis:7-alpine
    container_name: synor-compute-redis
    hostname: compute-redis
    restart: unless-stopped
    command: redis-server --appendonly yes --maxmemory 1gb --maxmemory-policy allkeys-lru
    ports:
      - "17280:6379"    # Redis port (remapped)
    networks:
      - synor-compute-net
    volumes:
      - compute-redis-data:/data
    healthcheck:
      test: ["CMD", "redis-cli", "ping"]
      interval: 15s
      timeout: 5s
      retries: 3

  # Prometheus metrics
  compute-prometheus:
    image: prom/prometheus:latest
    container_name: synor-compute-prometheus
    hostname: compute-prometheus
    restart: unless-stopped
    command:
      - '--config.file=/etc/prometheus/prometheus.yml'
      - '--storage.tsdb.path=/prometheus'
      - '--web.enable-lifecycle'
    volumes:
      - ./docker/compute-node/prometheus.yml:/etc/prometheus/prometheus.yml:ro
      - compute-prometheus-data:/prometheus
    ports:
      - "17290:9090"    # Prometheus UI (remapped)
    networks:
      - synor-compute-net
    healthcheck:
      test: ["CMD", "wget", "-q", "--spider", "http://localhost:9090/-/healthy"]
      interval: 30s
      timeout: 5s
      retries: 3

volumes:
  compute-orchestrator-data:
    driver: local
  compute-worker-cpu-1-data:
    driver: local
  compute-worker-cpu-2-data:
    driver: local
  compute-worker-wasm-data:
    driver: local
  compute-spot-market-data:
    driver: local
  compute-redis-data:
    driver: local
  compute-prometheus-data:
    driver: local

networks:
  synor-compute-net:
    driver: bridge
    ipam:
      config:
        - subnet: 172.23.0.0/16