Stirling-PDF/testing/compose/docker-compose-multinode.betterdb.yml

# BetterDB observability over the Valkey backplane plus an on-demand load generator. Dashboard: http://localhost:3001
# Up: docker compose -f docker-compose-multinode.yml -f docker-compose-multinode.betterdb.yml up -d   (load: append --profile load run --rm loadgen)

services:
  # Same Valkey, retuned so the observability panels have data on a local box.
  valkey:
    command:
      - valkey-server
      - --save
      - ""
      - --appendonly
      - "no"
      # Default 10000us never trips locally, so the slowlog stays empty; 500us shows real command spread.
      - --slowlog-log-slower-than
      - "500"
      - --slowlog-max-len
      - "1024"
      # 0 disables latency monitoring; 50ms populates LATENCY HISTORY/LATEST.
      - --latency-monitor-threshold
      - "50"
      - --maxmemory-policy
      - "noeviction"
    ports:
      - "6379:6379" # host access for valkey-cli inspection

  # One-shot: BetterDB runs as UID 1001 and cannot write a root-owned volume.
  betterdb-init:
    image: alpine:3.20
    container_name: multinode-betterdb-init
    command: ["sh", "-c", "chown -R 1001:1001 /data && echo 'betterdb data dir ready'"]
    volumes:
      - betterdb-data:/data
    networks:
      - stirling-multinode

  # One-shot: its own database on the stack's Postgres, so BetterDB history survives restarts.
  betterdb-db-init:
    image: postgres:17-alpine
    container_name: multinode-betterdb-db-init
    depends_on:
      postgres:
        condition: service_healthy
    environment:
      PGHOST: postgres
      PGUSER: stirling
      PGPASSWORD: stirling
      PGDATABASE: stirling
    command:
      - sh
      - -c
      - psql -tAc "SELECT 1 FROM pg_database WHERE datname='betterdb'" | grep -q 1 || psql -c "CREATE DATABASE betterdb"
    networks:
      - stirling-multinode

  betterdb:
    image: betterdb/monitor:latest
    container_name: multinode-betterdb
    restart: unless-stopped
    depends_on:
      valkey:
        condition: service_healthy
      betterdb-init:
        condition: service_completed_successfully
      betterdb-db-init:
        condition: service_completed_successfully
    environment:
      DB_HOST: valkey
      DB_PORT: "6379"
      DB_TYPE: valkey
      # postgres (not the default memory store) so history survives a restart; sqlite is not
      # compiled into the published image despite the docs listing it.
      STORAGE_TYPE: postgres
      STORAGE_URL: "postgresql://stirling:stirling@postgres:5432/betterdb"
      BETTERDB_DATA_DIR: /app/data
      # Encrypts stored connection secrets at rest; test-only value.
      ENCRYPTION_KEY: "multinode-betterdb-test-encryption-key"
      ANOMALY_DETECTION_ENABLED: "true"
      ANOMALY_POLL_INTERVAL_MS: "1000"
      AUDIT_POLL_INTERVAL_MS: "15000"
      CLIENT_ANALYTICS_POLL_INTERVAL_MS: "15000"
      KEY_ANALYTICS_INTERVAL_MS: "60000"
      # No phoning home from a local test stack.
      BETTERDB_TELEMETRY: "false"
    ports:
      - "3001:3001"
    volumes:
      - betterdb-data:/app/data
    # The image's own healthcheck probes "localhost", which resolves to ::1 while the server
    # binds IPv4 only, so it always reports unhealthy. Pin it to 127.0.0.1.
    healthcheck:
      test: ["CMD-SHELL", "wget -q -O- http://127.0.0.1:3001/api/health >/dev/null || exit 1"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 20s
    networks:
      - stirling-multinode

  # Profile-gated so `up` skips it; run on demand to drive traffic through the LB.
  loadgen:
    image: python:3.12-alpine
    container_name: multinode-loadgen
    profiles: ["load"]
    environment:
      BASE_URL: "http://nginx:8080"
      DURATION_SECONDS: "${DURATION_SECONDS:-300}"
      CONCURRENCY: "${CONCURRENCY:-24}"
      ASYNC_RATIO: "${ASYNC_RATIO:-0.35}"
      USER_COUNT: "${USER_COUNT:-40}"
    volumes:
      - ./multinode/loadgen.py:/loadgen.py:ro
    command: ["python3", "-u", "/loadgen.py"]
    networks:
      - stirling-multinode

volumes:
  betterdb-data: