docker-compose.yml

# docker-compose.yml — Phase 2.6: bootstrap + orchestrator only
#
# The orchestrator manages all pub-sub components (brokers, publishers, subscribers)
# dynamically via the Docker SDK. Static infrastructure only here.
#
# Start the full system:
#   make demo     # start, wait for orchestrator, seed 3+2+3 topology
#   make clean    # tear down everything (compose + dynamic containers)
#
# Networking note:
#   All containers join aether-net. The orchestrator creates dynamic containers
#   on this same network so they can resolve each other by hostname.

services:
  # ── bootstrap ─────────────────────────────────────────────────────────────
  bootstrap:
    build: .
    image: aether:latest
    container_name: aether-bootstrap
    hostname: bootstrap
    command: >
      aether-bootstrap
        --host bootstrap
        --port 7000
        --status-port 17000
        --log-level INFO
    environment:
      - AETHER_CONFIG=/app/config.docker.yaml
    volumes:
      - ./config.docker.yaml:/app/config.docker.yaml:ro
    ports:
      - "7100:7000"   # TCP peer-discovery (host 7100 → container 7000)
      - "17100:17000" # HTTP /status     (host 17100 → container 17000)
    healthcheck:
      test: ["CMD", "curl", "-sf", "http://localhost:17000/status"]
      interval: 5s
      timeout: 3s
      retries: 5
      start_period: 5s
    networks:
      - aether-net
    restart: "no"

  # ── orchestrator ──────────────────────────────────────────────────────────
  orchestrator:
    image: aether:latest
    container_name: aether-orchestrator
    hostname: orchestrator
    command: uvicorn aether.orchestrator.main:app --host 0.0.0.0 --port 9000 --log-level info
    ports:
      - "9000:9000"   # FastAPI control plane
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock  # Docker-out-of-Docker
    user: root  # required to access the Docker socket
    environment:
      BOOTSTRAP_HOST: bootstrap
      BOOTSTRAP_PORT: "7000"
      AETHER_IMAGE: aether:latest
      DOCKER_NETWORK: aether-net
      OTEL_ENDPOINT: http://otel-collector:4318
    depends_on:
      bootstrap:
        condition: service_healthy
    healthcheck:
      test: ["CMD", "curl", "-sf", "http://localhost:9000/docs"]
      interval: 5s
      timeout: 3s
      retries: 10
      start_period: 15s
    networks:
      - aether-net
    restart: "no"

  # ── dashboard ──────────────────────────────────────────────────────────────
  dashboard:
    build:
      context: ./dashboard
      dockerfile: Dockerfile
    image: aether-dashboard:latest
    container_name: aether-dashboard
    hostname: dashboard
    ports:
      - "3000:80"    # React dashboard (host 3000 → nginx 80)
    depends_on:
      orchestrator:
        condition: service_healthy
    healthcheck:
      test: ["CMD", "wget", "-qO-", "http://127.0.0.1:80/"]
      interval: 10s
      timeout: 5s
      retries: 3
      start_period: 10s
    networks:
      - aether-net
    restart: "no"

  # ── otel-collector ────────────────────────────────────────────────────────
  # Receives OTLP logs from all Aether processes and forwards to Loki.
  # Pinned to a specific version — the Loki exporter config schema changes
  # between otelcol-contrib releases and latest can break the config silently.
  otel-collector:
    image: otel/opentelemetry-collector-contrib:0.95.0
    container_name: aether-otel-collector
    hostname: otel-collector
    command: ["--config=/etc/otel-collector.yaml"]
    volumes:
      - ./observability/otel-collector.yaml:/etc/otel-collector.yaml:ro
    ports:
      - "4317:4317"   # OTLP gRPC (future use)
      - "4318:4318"   # OTLP HTTP (used by Python opentelemetry-exporter-otlp-proto-http)
    depends_on:
      - loki
    networks:
      - aether-net
    restart: unless-stopped

  # ── loki ──────────────────────────────────────────────────────────────────
  # Log storage and query engine. Receives from otel-collector, queried by Grafana.
  loki:
    image: grafana/loki:2.9.4
    container_name: aether-loki
    hostname: loki
    command: -config.file=/etc/loki/loki.yaml
    volumes:
      - ./observability/loki.yaml:/etc/loki/loki.yaml:ro
    ports:
      - "3100:3100"
    networks:
      - aether-net
    restart: unless-stopped

  # ── prometheus ────────────────────────────────────────────────────────────
  # Scrapes /metrics from the orchestrator every 15s. No hard startup deps —
  # Prometheus retries scrape targets that aren't ready yet.
  prometheus:
    image: prom/prometheus:v2.50.1
    container_name: aether-prometheus
    hostname: prometheus
    volumes:
      - ./observability/prometheus.yaml:/etc/prometheus/prometheus.yml:ro
    ports:
      - "9090:9090"
    networks:
      - aether-net
    restart: unless-stopped

  # ── grafana ───────────────────────────────────────────────────────────────
  # Dashboards and log/metric exploration. Available at http://localhost:3001
  # (port 3000 is taken by the React dashboard). Anonymous access enabled so
  # there's no login step in the demo.
  # Datasources (Prometheus + Loki) and dashboard provider are provisioned
  # automatically from observability/grafana/provisioning/ on first boot.
  grafana:
    image: grafana/grafana:10.3.3
    container_name: aether-grafana
    hostname: grafana
    environment:
      GF_SERVER_HTTP_PORT: "3001"
      GF_AUTH_ANONYMOUS_ENABLED: "true"
      GF_AUTH_ANONYMOUS_ORG_ROLE: "Viewer"
    volumes:
      - ./observability/grafana/provisioning:/etc/grafana/provisioning:ro
      - ./observability/grafana/dashboards:/var/lib/grafana/dashboards:ro
    ports:
      - "3001:3001"
    depends_on:
      - loki
      - prometheus
    networks:
      - aether-net
    restart: unless-stopped

# ── network ───────────────────────────────────────────────────────────────
networks:
  aether-net:
    name: aether-net
    driver: bridge