docker-compose.yml

# PayFlow — Distributed Payment Processing System
# Docker Compose Orchestration for all services
# Owner: Member 5 (Monitoring, Testing & Infrastructure)
#
# Usage:
#   docker compose up --build        — start all services
#   docker compose down --volumes    — stop and clean up
#   docker compose logs <service>    — view service logs

x-coordinator-base: &coordinator-base
  build:
    context: ./payflow
    dockerfile: coordinator/Dockerfile
  cap_add:
    - NET_ADMIN
  deploy:
    resources:
      limits:
        cpus: "0.5"
        memory: "256M"
  restart: unless-stopped
  networks:
    - payflow-net
  depends_on:
    payment-log:
      condition: service_healthy
  healthcheck:
    interval: 10s
    timeout: 5s
    retries: 3
    start_period: 20s

x-worker-base: &worker-base
  build:
    context: ./payflow/worker
    dockerfile: Dockerfile
  cap_add:
    - NET_ADMIN
  deploy:
    resources:
      limits:
        cpus: "0.5"
        memory: "128M"
  restart: unless-stopped
  networks:
    - payflow-net
  depends_on:
    coordinator-1:
      condition: service_healthy
    coordinator-2:
      condition: service_healthy
    coordinator-3:
      condition: service_healthy
  healthcheck:
    test: ["CMD", "wget", "-qO-", "http://localhost:2112/metrics"]
    interval: 10s
    timeout: 5s
    retries: 3
    start_period: 15s

x-worker-env: &worker-env
  METRICS_PORT: "2112"
  COORDINATOR_ADDR: "coordinator-3:50053"
  LOG_SERVICE_ADDR: "payment-log:50054"
  HEARTBEAT_INTERVAL: "2s"
  BANK_API_ADDR: "http://mock-bank:9999"
  BANK_LATENCY_MIN_MS: "50"
  BANK_LATENCY_MAX_MS: "500"
  BANK_FAIL_RATE: "0.10"
  WORKER_TOKEN: "secret-token"
  OUTBOX_DB_PATH: "/tmp/outbox"
  OTEL_EXPORTER_OTLP_ENDPOINT: "http://jaeger:4318"

services:
  # ─────────────────────────────────────────────────────────
  # C1: API Gateway — REST + gRPC entry point
  # ─────────────────────────────────────────────────────────
  api-gateway:
    build:
      context: ./payflow
      dockerfile: gateway/Dockerfile
    cap_add:
      - NET_ADMIN
    ports:
      - "8080:8080"
      - "9090:9090"
    environment:
      PORT: "8080"
      METRICS_PORT: "2112"
      GRPC_PORT: "9090"
      JWT_SECRET: "dev-secret-change-in-prod"
      COORDINATOR_ADDRS: "coordinator-1:50051,coordinator-2:50052,coordinator-3:50053"
      OTEL_EXPORTER_OTLP_ENDPOINT: "http://jaeger:4318"
      OTEL_SERVICE_NAME: "api-gateway"
    healthcheck:
      test: ["CMD", "wget", "-qO-", "http://localhost:8080/health"]
      interval: 10s
      timeout: 5s
      retries: 3
      start_period: 15s
    depends_on:
      coordinator-1:
        condition: service_healthy
      coordinator-2:
        condition: service_healthy
      coordinator-3:
        condition: service_healthy
    deploy:
      resources:
        limits:
          cpus: "0.5"
          memory: "256M"
    restart: unless-stopped
    networks:
      - payflow-net

  # ─────────────────────────────────────────────────────────
  # C2: Coordinator Cluster — Bully Election + Task Dispatch
  # ─────────────────────────────────────────────────────────
  coordinator-1:
    <<: *coordinator-base
    command: ["--id=coordinator-1", "--port=:50051"]
    ports:
      - "50051:50051"
      - "2112:2112"
    environment:
      NODE_ID: "1"
      PORT: "50051"
      GRPC_PORT: "50051"
      METRICS_PORT: "2112"
      PEERS: "coordinator-2:50052,coordinator-3:50053"
      PAYMENT_LOG_ADDR: "payment-log:50054"
      HEARTBEAT_INTERVAL: "2s"
      ELECTION_TIMEOUT: "5s"
      ELECTION_TIMEOUT_MS: "5000"
      HEARTBEAT_INTERVAL_MS: "2000"
      HEARTBEAT_MISS_THRESHOLD: "3"
      OTEL_EXPORTER_OTLP_ENDPOINT: "http://jaeger:4318"
      OTEL_SERVICE_NAME: "coordinator-1"
    healthcheck:
      test: ["CMD-SHELL", "nc -z localhost 50051 || exit 1"]

  coordinator-2:
    <<: *coordinator-base
    command: ["--id=coordinator-2", "--port=:50052"]
    ports:
      - "50052:50052"
      - "2153:2112"
    environment:
      NODE_ID: "2"
      PORT: "50052"
      GRPC_PORT: "50052"
      METRICS_PORT: "2112"
      PEERS: "coordinator-1:50051,coordinator-3:50053"
      PAYMENT_LOG_ADDR: "payment-log:50054"
      HEARTBEAT_INTERVAL: "2s"
      ELECTION_TIMEOUT: "5s"
      ELECTION_TIMEOUT_MS: "5000"
      HEARTBEAT_INTERVAL_MS: "2000"
      HEARTBEAT_MISS_THRESHOLD: "3"
      OTEL_EXPORTER_OTLP_ENDPOINT: "http://jaeger:4318"
      OTEL_SERVICE_NAME: "coordinator-2"
    healthcheck:
      test: ["CMD-SHELL", "nc -z localhost 50052 || exit 1"]

  coordinator-3:
    <<: *coordinator-base
    command: ["--id=coordinator-3", "--port=:50053"]
    ports:
      - "50053:50053"
      - "2154:2112"
    environment:
      NODE_ID: "3"
      PORT: "50053"
      GRPC_PORT: "50053"
      METRICS_PORT: "2112"
      PEERS: "coordinator-1:50051,coordinator-2:50052"
      PAYMENT_LOG_ADDR: "payment-log:50054"
      HEARTBEAT_INTERVAL: "2s"
      ELECTION_TIMEOUT: "5s"
      ELECTION_TIMEOUT_MS: "5000"
      HEARTBEAT_INTERVAL_MS: "2000"
      HEARTBEAT_MISS_THRESHOLD: "3"
      OTEL_EXPORTER_OTLP_ENDPOINT: "http://jaeger:4318"
      OTEL_SERVICE_NAME: "coordinator-3"
    healthcheck:
      test: ["CMD-SHELL", "nc -z localhost 50053 || exit 1"]

  # ─────────────────────────────────────────────────────────
  # C3: Worker Services — Payment Execution + Heartbeat
  # ─────────────────────────────────────────────────────────
  worker-1:
    <<: *worker-base
    environment:
      <<: *worker-env
      WORKER_ID: "1"
      PORT: "8080"
      OTEL_SERVICE_NAME: "worker-1"

  worker-2:
    <<: *worker-base
    environment:
      <<: *worker-env
      WORKER_ID: "2"
      PORT: "8080"
      OTEL_SERVICE_NAME: "worker-2"

  worker-3:
    <<: *worker-base
    environment:
      <<: *worker-env
      WORKER_ID: "3"
      PORT: "8080"
      OTEL_SERVICE_NAME: "worker-3"

  worker-4:
    <<: *worker-base
    environment:
      <<: *worker-env
      WORKER_ID: "4"
      PORT: "8080"
      OTEL_SERVICE_NAME: "worker-4"

  worker-5:
    <<: *worker-base
    environment:
      <<: *worker-env
      WORKER_ID: "5"
      PORT: "8080"
      OTEL_SERVICE_NAME: "worker-5"

  # ─────────────────────────────────────────────────────────
  # C4: Payment Log Service — BoltDB append-only log
  # ─────────────────────────────────────────────────────────
  payment-log:
    build:
      context: ./payflow
      dockerfile: payment-log/Dockerfile
    cap_add:
      - NET_ADMIN
    ports:
      - "50054:50054"
      - "2113:2113"
    environment:
      PORT: "2113"
      GRPC_PORT: "50054"
      METRICS_PORT: "2112"
      HEALTH_PORT: "2113"
      BOLT_PATH: "/data/payments.db"
      OTEL_EXPORTER_OTLP_ENDPOINT: "http://jaeger:4318"
      OTEL_SERVICE_NAME: "payment-log"
    volumes:
      - payflow-bolt:/data
    healthcheck:
      test: ["CMD-SHELL", "nc -z localhost 50054 || exit 1"]
      interval: 10s
      timeout: 5s
      retries: 3
      start_period: 20s
    deploy:
      resources:
        limits:
          cpus: "1.0"
          memory: "512M"
    restart: unless-stopped
    networks:
      - payflow-net

  # ─────────────────────────────────────────────────────────
  # C5: Monitoring Service — Prometheus + WebSocket Dashboard
  # ─────────────────────────────────────────────────────────
  monitor:
    build:
      context: ./payflow/monitor
    cap_add:
      - NET_ADMIN
    ports:
      - "3000:3000"
      - "9091:9091"
    environment:
      HTTP_PORT: "3000"
      PROMETHEUS_PORT: "9091"
      SCRAPE_INTERVAL: "15s"
      SCALING_ENABLED: "true"
      SCALING_QUEUE_THRESHOLD: "50"
      SCALING_MAX_WORKERS: "10"
      SCALING_COOLDOWN: "30s"
      SCALING_TEMPLATE_WORKER: "worker-1"
      SCRAPE_TARGETS: >-
        http://api-gateway:8080/metrics,
        http://coordinator-1:2112/metrics,
        http://coordinator-2:2112/metrics,
        http://coordinator-3:2112/metrics,
        http://payment-log:2112/metrics,
        http://worker-1:2112/metrics,
        http://worker-2:2112/metrics,
        http://worker-3:2112/metrics,
        http://worker-4:2112/metrics,
        http://worker-5:2112/metrics
      OTEL_EXPORTER_OTLP_ENDPOINT: "http://jaeger:4318"
      OTEL_SERVICE_NAME: "monitor"
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock
    healthcheck:
      test: ["CMD", "wget", "-qO-", "http://localhost:3000/health"]
      interval: 10s
      timeout: 5s
      retries: 3
      start_period: 15s
    depends_on:
      api-gateway:
        condition: service_healthy
    deploy:
      resources:
        limits:
          cpus: "0.5"
          memory: "256M"
    restart: unless-stopped
    networks:
      - payflow-net

  # ─────────────────────────────────────────────────────────
  # Supporting: Mock Bank API (simulates external payment gateway)
  # Uses placeholder service - workers call /v1/payments endpoint
  # ─────────────────────────────────────────────────────────
  mock-bank:
    build:
      context: ./payflow/placeholder
      dockerfile: Dockerfile
    ports:
      - "9999:9999"
    environment:
      PORT: "9999"
      METRICS_PORT: "2113"
    healthcheck:
      test: ["CMD", "wget", "-qO-", "http://localhost:9999/health"]
      interval: 10s
      timeout: 5s
      retries: 3
      start_period: 10s
    deploy:
      resources:
        limits:
          cpus: "0.25"
          memory: "64M"
    restart: unless-stopped
    networks:
      - payflow-net

  # ─────────────────────────────────────────────────────────
  # Supporting: Jaeger Distributed Tracing
  # ─────────────────────────────────────────────────────────
  jaeger:
    image: jaegertracing/all-in-one:1.55
    ports:
      - "16686:16686"
      - "14268:14268"
      - "4318:4318"
    environment:
      COLLECTOR_OTLP_ENABLED: "true"
    healthcheck:
      test: ["CMD", "wget", "-qO-", "http://localhost:16686/"]
      interval: 10s
      timeout: 5s
      retries: 3
      start_period: 20s
    deploy:
      resources:
        limits:
          cpus: "0.5"
          memory: "256M"
    restart: unless-stopped
    networks:
      - payflow-net

volumes:
  payflow-bolt:
    driver: local
    labels:
      com.payflow.description: "BoltDB persistent store for C4 payment-log"

networks:
  payflow-net:
    driver: bridge
    ipam:
      config:
        - subnet: 172.20.0.0/16