docker-compose.bench.yml

name: relier-bench

# =============================================================================
# Isolated benchmark environment — runs entirely inside Docker so results
# are not distorted by Windows-only solo pool or by competing dev workers.
#
# Default run (500 tasks, synthetic 0.5 s tasks):
#   docker compose -f docker-compose.bench.yml up --build
#
# Scale to 10k tasks:
#   BENCH_BATCH_SIZE=10000 docker compose -f docker-compose.bench.yml up --build
#
# Scale to 100k tasks (add more worker replicas first):
#   BENCH_BATCH_SIZE=100000 BENCH_WORKER_CONCURRENCY=8 \
#     docker compose -f docker-compose.bench.yml up --build --scale bench-runner=1
#
# Grafana: http://localhost:3001  (user: admin / bench)
# Prometheus: http://localhost:9091
# Redis (isolated): localhost:6380
# =============================================================================

x-bench-env: &bench-env
  RELIER_REDIS_URL: redis://bench-redis:6379/0
  BENCH_SYNTHETIC: "1"
  BENCH_SYNTHETIC_SLEEP: "${BENCH_SYNTHETIC_SLEEP:-0.5}"
  BENCH_BATCH_SIZE: "${BENCH_BATCH_SIZE:-500}"
  # Disable OTel in bench the bench measures wall-clock time, not OTEL traces.
  RELIER_OTEL_ENABLED: "false"

services:
  # ===========================================================================
  # Infrastructure — isolated Redis, never shared with the dev stack
  # ===========================================================================
  bench-redis:
    image: redis:7.2-alpine
    container_name: bench-redis
    # Minimal inline config: noeviction + AOF, no external redis.conf needed.
    command: >
      redis-server
      --maxmemory-policy noeviction
      --appendonly yes
      --save 60 1
    ports:
      - "6380:6379"
    healthcheck:
      test: ["CMD", "redis-cli", "ping"]
      interval: 3s
      timeout: 2s
      retries: 15
    networks:
      - bench-net

  # ===========================================================================
  # Observability
  # ===========================================================================
  redis-exporter:
    image: oliver006/redis_exporter:latest
    container_name: bench-redis-exporter
    command:
      - "--redis.addr=redis://bench-redis:6379"
      - "--check-keys=default,high_priority,low_priority,re-queue,vanilla,vanilla_acks_late,rl:m:global:success,rl:m:global:failed,rl:m:global:resurrected,bench:relier:embed_exec,bench:relier:noop_exec,bench:relier:summary_exec,bench:relier:delivery_done,bench:relier:oom_done,bench:relier:oom_exec,bench:vanilla:embed_exec,bench:vanilla:noop_exec,bench:vanilla:delivery_done,bench:vanilla:oom_done,bench:vanilla_acks_late:delivery_done,bench:vanilla_acks_late:total_exec,bench:relier:phoenix_load_started,bench:relier:phoenix_load_done,bench:cold_start:p50_ms,bench:cold_start:p99_ms"
    ports:
      - "9121:9121"
    depends_on:
      bench-redis:
        condition: service_healthy
    networks:
      - bench-net

  prometheus:
    image: prom/prometheus:v2.52.0
    container_name: bench-prometheus
    command:
      - "--config.file=/etc/prometheus/prometheus.yml"
      - "--storage.tsdb.path=/prometheus"
      - "--storage.tsdb.retention.time=2h"
      - "--web.enable-lifecycle"
    volumes:
      - ./bench/prometheus.yml:/etc/prometheus/prometheus.yml:ro
      - bench-prometheus-data:/prometheus
    ports:
      - "9091:9090"
    depends_on:
      - redis-exporter
    networks:
      - bench-net

  grafana:
    image: grafana/grafana:11.0.0
    container_name: bench-grafana
    environment:
      GF_SECURITY_ADMIN_PASSWORD: bench
      GF_AUTH_ANONYMOUS_ENABLED: "true"
      GF_AUTH_ANONYMOUS_ORG_ROLE: Viewer
      GF_USERS_DEFAULT_THEME: dark
      GF_DASHBOARDS_DEFAULT_HOME_DASHBOARD_PATH: /var/lib/grafana/dashboards/bench-overview.json
    volumes:
      - ./bench/grafana/provisioning:/etc/grafana/provisioning:ro
      - ./bench/grafana/dashboards:/var/lib/grafana/dashboards:ro
      - bench-grafana-data:/var/lib/grafana
    ports:
      # Port 3001 avoids conflicting with dev Grafana on 3000.
      - "3001:3000"
    depends_on:
      - prometheus
    networks:
      - bench-net

  # ===========================================================================
  # Bench runner — manages all worker subprocesses internally.
  #
  # The bench spawns, kills, and replaces workers as part of its test protocol
  # (OOM simulation, delivery-rate crash testing). All subprocesses run inside
  # this container.
  # ===========================================================================
  bench-runner:
    build:
      context: .
      dockerfile: Dockerfile.bench
    container_name: bench-runner
    command: python -m bench.bench --synthetic
    environment:
      <<: *bench-env
    volumes:
      # Mount bench source live so you can edit bench tasks without rebuilding.
      - ./bench:/app/bench:ro
      - ./src:/app/src:ro
      # Results land here inspect on the host after the run.
      - ./bench-results:/app/bench-results
    depends_on:
      bench-redis:
        condition: service_healthy
    networks:
      - bench-net
    # bench-runner exits when the bench finishes; don't restart it.
    restart: "no"

networks:
  bench-net:
    driver: bridge

volumes:
  bench-prometheus-data:
  bench-grafana-data: