docker-compose.yml

# Phase-5 observability: labels + json-file logging anchors.
# HSH-HQ Alloy discovers this container by `com.barnardhq.*` and stamps
# project/service/tenant/env on every log stream. `docker-compose.demo.yml`
# (CHAD-HQ) overrides `com.barnardhq.env` to `demo`. Do NOT change label
# keys without updating `/opt/observability-agent/alloy/config.alloy`.
x-barnardhq-labels: &barnardhq-labels
  com.barnardhq.project: "droneops"
  com.barnardhq.env: "prod"
  com.barnardhq.tenant: "shared"
  com.barnardhq.stack: "droneops"

x-logging: &logging
  driver: json-file
  options:
    max-size: "50m"
    max-file: "5"
    labels: "com.barnardhq.service,com.barnardhq.project,com.barnardhq.env,com.barnardhq.tenant,com.barnardhq.stack"

services:
  db:
    image: postgres:16-alpine
    restart: unless-stopped
    entrypoint: ["/scripts/primary-entrypoint.sh"]
    command: ["postgres"]
    environment:
      POSTGRES_USER: ${POSTGRES_USER:-doc}
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-changeme_in_production}
      POSTGRES_DB: ${POSTGRES_DB:-doc}
      REPLICATION_PASSWORD: ${REPLICATION_PASSWORD:-SecureDroneRepl2026}
    volumes:
      - postgres_data:/var/lib/postgresql/data
      - ./scripts/primary-entrypoint.sh:/scripts/primary-entrypoint.sh:ro
      - ./scripts/init-primary.sh:/docker-entrypoint-initdb.d/init-primary.sh:ro
    ports:
      - "127.0.0.1:5434:5432"
      - "10.99.0.1:5434:5432"
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U ${POSTGRES_USER:-doc}"]
      interval: 5s
      timeout: 5s
      retries: 5
    labels:
      <<: *barnardhq-labels
      com.barnardhq.service: "droneops-db"
    logging: *logging

  redis:
    image: redis:7-alpine
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "redis-cli", "ping"]
      interval: 5s
      timeout: 5s
      retries: 5
    labels:
      <<: *barnardhq-labels
      com.barnardhq.service: "droneops-redis"
    logging: *logging

  ollama:
    image: ollama/ollama:latest
    restart: unless-stopped
    volumes:
      - ollama_data:/root/.ollama
    environment:
      - OLLAMA_HOST=0.0.0.0
      # Keep model loaded in RAM permanently
      - OLLAMA_KEEP_ALIVE=-1
      # Single-request mode — all CPU goes to one generation
      - OLLAMA_NUM_PARALLEL=1
      # Only keep one model loaded (saves RAM for the model itself)
      - OLLAMA_MAX_LOADED_MODELS=1
      # Enable flash attention for faster inference
      - OLLAMA_FLASH_ATTENTION=1
    cpuset: "0-5"
    mem_reservation: 8g
    healthcheck:
      test: ["CMD-SHELL", "bash -c 'echo > /dev/tcp/localhost/11434' || exit 1"]
      interval: 15s
      timeout: 10s
      retries: 5
      start_period: 30s
    labels:
      <<: *barnardhq-labels
      com.barnardhq.service: "droneops-ollama"
    logging: *logging

  ollama-setup:
    image: curlimages/curl:latest
    depends_on:
      ollama:
        condition: service_healthy
    restart: "no"
    entrypoint: >
      sh -c "
        echo 'Waiting for Ollama to be ready...' &&
        until curl -s http://ollama:11434/api/tags > /dev/null 2>&1; do sleep 2; done &&
        echo 'Pulling Llama 3.1 8B model...' &&
        curl -s http://ollama:11434/api/pull -d '{\"name\": \"llama3.1:8b-instruct-q4_K_M\"}' &&
        echo 'Model pulled. Will load into RAM on first use.'
      "
    labels:
      <<: *barnardhq-labels
      com.barnardhq.service: "droneops-ollama-setup"
    logging: *logging

  # ── Watchtower — Auto-update monitoring ────────────────────────
  # Checks for new images daily and sends notifications.
  # Set WATCHTOWER_NOTIFICATION_URL in .env for alerts.
  watchtower:
    image: containrrr/watchtower:latest
    restart: unless-stopped
    environment:
      - WATCHTOWER_CLEANUP=true
      - WATCHTOWER_LABEL_ENABLE=true
      - WATCHTOWER_POLL_INTERVAL=86400
      - WATCHTOWER_MONITOR_ONLY=${WATCHTOWER_MONITOR_ONLY:-false}
      - WATCHTOWER_NOTIFICATION_URL=${WATCHTOWER_NOTIFICATION_URL:-}
      - WATCHTOWER_NOTIFICATIONS_HOSTNAME=droneops-server
      - DOCKER_API_VERSION=1.45
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock
    labels:
      <<: *barnardhq-labels
      com.barnardhq.service: "droneops-watchtower"
    logging: *logging

  backend:
    build:
      context: ./backend
      dockerfile: Dockerfile
    restart: unless-stopped
    # tini reaps orphan curl/subprocesses spawned by FastAPI handlers so
    # they never accumulate as <defunct> (matches worker service; observed
    # uvicorn zombies 2026-04-19 during HSH-HQ high-load incident).
    init: true
    dns:
      - 1.1.1.1
      - 8.8.8.8
    environment:
      - DATABASE_URL=${DATABASE_URL:-postgresql+asyncpg://doc:changeme_in_production@db:5432/doc}
      - REDIS_URL=${REDIS_URL:-redis://redis:6379/0}
      - OLLAMA_BASE_URL=${OLLAMA_BASE_URL:-http://ollama:11434}
      - OLLAMA_MODEL=${OLLAMA_MODEL:-llama3.1:8b-instruct-q4_K_M}
      - OPENDRONELOG_URL=${OPENDRONELOG_URL:-}
      - JWT_SECRET_KEY=${JWT_SECRET_KEY:-changeme_generate_a_random_secret}
      - JWT_ALGORITHM=${JWT_ALGORITHM:-HS256}
      - JWT_ACCESS_TOKEN_EXPIRE_MINUTES=${JWT_ACCESS_TOKEN_EXPIRE_MINUTES:-30}
      - JWT_REFRESH_TOKEN_EXPIRE_DAYS=${JWT_REFRESH_TOKEN_EXPIRE_DAYS:-30}
      - DEMO_ADMIN_USERNAME=${DEMO_ADMIN_USERNAME:-}
      - DEMO_ADMIN_PASSWORD=${DEMO_ADMIN_PASSWORD:-}
      - SMTP_HOST=${SMTP_HOST:-}
      - SMTP_PORT=${SMTP_PORT:-587}
      - SMTP_USER=${SMTP_USER:-}
      - SMTP_PASSWORD=${SMTP_PASSWORD:-}
      - SMTP_FROM_EMAIL=${SMTP_FROM_EMAIL:-}
      - SMTP_FROM_NAME=${SMTP_FROM_NAME:-}
      - SMTP_USE_TLS=${SMTP_USE_TLS:-true}
      - UPLOAD_DIR=/data/uploads
      - REPORTS_DIR=/data/reports
      - FRONTEND_URL=${FRONTEND_URL:-http://localhost:3080}
      # Phase-5 observability — all DSN/endpoint gated (unset = no-op).
      - SENTRY_DSN=${SENTRY_DSN:-}
      - SENTRY_TRACES_SAMPLE_RATE=${SENTRY_TRACES_SAMPLE_RATE:-0.05}
      - SENTRY_ENVIRONMENT=${SENTRY_ENVIRONMENT:-production}
      - OTEL_EXPORTER_OTLP_ENDPOINT=${OTEL_EXPORTER_OTLP_ENDPOINT:-}
      - OTEL_SERVICE_NAME=droneops-api
      - TENANT=${TENANT:-shared}
      - ENV=${ENV:-prod}
      - APP_VERSION=2.63.1
    volumes:
      - app_data:/data
    ports:
      - "8000:8000"
    depends_on:
      db:
        condition: service_healthy
      redis:
        condition: service_healthy
      ollama:
        condition: service_healthy
    healthcheck:
      test: ["CMD-SHELL", "curl -sf http://localhost:8000/api/health || exit 1"]
      interval: 15s
      timeout: 10s
      retries: 5
      start_period: 45s
    labels:
      <<: *barnardhq-labels
      com.barnardhq.service: "droneops-api"
    logging: *logging

  worker:
    build:
      context: ./backend
      dockerfile: Dockerfile
    restart: unless-stopped
    # PID 1 init reaper — without this, celery prefork children that exit
    # become zombies if the master loses a SIGCHLD reap. The 2026-04-19
    # high-load incident found 33 defunct celery children of the worker
    # master accumulating ~2/hr over 18h, contributing to the
    # HostZombieProcesses alert. tini (docker init: true) reaps any orphan.
    # See docs/incidents/2026-04-19-hsh-hq-high-load.md.
    init: true
    # --max-tasks-per-child=50 forces each prefork child to recycle after
    # 50 tasks. Combined with init: true, this guarantees zombies cannot
    # accumulate even if a child cleanup path raises. 50 is conservative —
    # report+email tasks are short-lived, child startup cost is dominated
    # by Sentry/OTEL init (~3-5s), so 50/child amortizes that fine.
    command: celery -A app.tasks.celery_tasks worker --loglevel=info --max-tasks-per-child=50
    healthcheck:
      # Redis-backed worker heartbeat (replaces the old `celery inspect ping`
      # subprocess which re-imported the full OTel chain every 60s — 3-5s of
      # wasted CPU per check). The worker writes a unix-ts key to Redis on
      # every `worker_heartbeat` Celery signal (~2s cadence). Healthcheck
      # reads the key and verifies age < 60s.
      #   - Fresh key → Celery control loop is alive → worker healthy.
      #   - Stale/missing key → worker frozen/crashed → docker restarts.
      # Fast path: no Python process spawn, just a Redis GET.
      test:
        - CMD-SHELL
        - >
          K=$$(redis-cli -h redis -p 6379 get droneops:worker:heartbeat 2>/dev/null);
          if [ -z "$$K" ]; then exit 1; fi;
          AGE=$$(( $$(date +%s) - $$K ));
          if [ $$AGE -gt 60 ]; then exit 1; else exit 0; fi
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 30s
    environment:
      - DATABASE_URL=${DATABASE_URL:-postgresql+asyncpg://doc:changeme_in_production@db:5432/doc}
      - REDIS_URL=${REDIS_URL:-redis://redis:6379/0}
      - OLLAMA_BASE_URL=${OLLAMA_BASE_URL:-http://ollama:11434}
      - OLLAMA_MODEL=${OLLAMA_MODEL:-llama3.1:8b-instruct-q4_K_M}
      - SMTP_HOST=${SMTP_HOST:-}
      - SMTP_PORT=${SMTP_PORT:-587}
      - SMTP_USER=${SMTP_USER:-}
      - SMTP_PASSWORD=${SMTP_PASSWORD:-}
      - SMTP_FROM_EMAIL=${SMTP_FROM_EMAIL:-}
      - SMTP_FROM_NAME=${SMTP_FROM_NAME:-}
      - SMTP_USE_TLS=${SMTP_USE_TLS:-true}
      - UPLOAD_DIR=/data/uploads
      - REPORTS_DIR=/data/reports
      # Phase-5 observability — all DSN/endpoint gated (unset = no-op).
      - SENTRY_DSN=${SENTRY_DSN:-}
      - SENTRY_TRACES_SAMPLE_RATE=${SENTRY_TRACES_SAMPLE_RATE:-0.05}
      - SENTRY_ENVIRONMENT=${SENTRY_ENVIRONMENT:-production}
      - OTEL_EXPORTER_OTLP_ENDPOINT=${OTEL_EXPORTER_OTLP_ENDPOINT:-}
      - OTEL_SERVICE_NAME=droneops-worker
      - TENANT=${TENANT:-shared}
      - ENV=${ENV:-prod}
      - APP_VERSION=2.63.1
    volumes:
      - app_data:/data
    depends_on:
      db:
        condition: service_healthy
      redis:
        condition: service_healthy
    labels:
      <<: *barnardhq-labels
      com.barnardhq.service: "droneops-worker"
    logging: *logging

  flight-parser:
    build:
      context: ./flight-parser
      dockerfile: Dockerfile
    restart: unless-stopped
    healthcheck:
      test: ["CMD-SHELL", "curl -sf http://localhost:8100/health || exit 1"]
      interval: 15s
      timeout: 10s
      retries: 3
      start_period: 15s
    dns:
      - 1.1.1.1
      - 8.8.8.8
    environment:
      - DJI_API_KEY=${DJI_API_KEY:-}
      - PARSER_PORT=8100
    labels:
      <<: *barnardhq-labels
      com.barnardhq.service: "droneops-flight-parser"
    logging: *logging

  frontend:
    build:
      context: ./frontend
      dockerfile: Dockerfile
      args:
        # Sentry DSN is bundled into the JS at build time — Vite inlines
        # any `VITE_*` var that's in scope when `npm run build` runs.
        VITE_SENTRY_DSN: ${VITE_SENTRY_DSN:-}
        VITE_SENTRY_ENVIRONMENT: ${VITE_SENTRY_ENVIRONMENT:-production}
        VITE_APP_VERSION: "2.63.1"
    restart: unless-stopped
    healthcheck:
      test: ["CMD-SHELL", "curl -sf http://localhost:8080/ || exit 1"]
      interval: 15s
      timeout: 10s
      retries: 3
      start_period: 10s
    ports:
      - "${FRONTEND_PORT:-3080}:8080"
    depends_on:
      backend:
        condition: service_healthy
    labels:
      <<: *barnardhq-labels
      com.barnardhq.service: "droneops-frontend"
    logging: *logging

  # ── Cloudflare Tunnel ─────────────────────────────────────────────
  # Exposes the app securely without opening router ports.
  # Set CLOUDFLARE_TUNNEL_TOKEN in .env — service is a no-op without it.
  cloudflared:
    image: cloudflare/cloudflared:latest
    restart: unless-stopped
    command: tunnel run
    environment:
      - TUNNEL_TOKEN=${CLOUDFLARE_TUNNEL_TOKEN:-}
    depends_on:
      frontend:
        condition: service_healthy
      backend:
        condition: service_healthy
    labels:
      <<: *barnardhq-labels
      com.barnardhq.service: "droneops-cloudflared"
    logging: *logging

volumes:
  postgres_data:
  ollama_data:
  app_data: