docker-compose.lite.yml

Rama main · réplica de lectura del repositorio original.

# Lightweight local dev stack. `just cluster` layers this on the dev profile by
# default; `just cluster up -d --full` runs every service instead.
#   docker compose -f docker-compose.dev.yml -f docker-compose.lite.yml up -d
#
# What changes:
# - Temporal runs as the single-binary dev server (SQLite, built-in UI on
#   TEMPORAL_UI_PORT) instead of auto-setup + its own Postgres + a UI container.
# - Optional services only start when their profile is enabled:
#   just cluster up -d --with agents,mcp   (or COMPOSE_PROFILES=agents,mcp)
#     agents          litellm, agent-worker, agent-executor
#     mcp             mcp
#   The background worker always runs: Git sync and semantic-search indexing.
# - Healthchecks probe every 1-2s while starting, then every 30s.
# - Memory caps per service, and MALLOC_ARENA_MAX=2 for Python services.
#
# Temporal history lives in its own volume (temporal-lite-data), so workflow
# runs from the full stack's Temporal are not visible here and vice versa.
# Tracecat data in Postgres, MinIO, and Redis is shared between modes.

x-temporal-image: &temporal-image temporalio/temporal:1.8.0

x-python-env: &python-env
  MALLOC_ARENA_MAX: ${MALLOC_ARENA_MAX:-2}

services:
  # The dev server runs as uid 1000, but a fresh named volume is root-owned.
  temporal-lite-init:
    image: *temporal-image
    user: "0:0"
    entrypoint: ["chown", "-R", "1000:1000", "/data"]
    volumes:
      - temporal-lite-data:/data
    network_mode: none
    restart: "no"
    attach: false

  temporal:
    image: *temporal-image
    # Pin the image's non-root user; temporal-lite-init chowns /data to match.
    user: "1000:1000"
    command:
      - server
      - start-dev
      - --ip=0.0.0.0
      - --db-filename=/data/temporal.db
      - --namespace=default
      - --ui-port=8080
      - --log-level=warn
    environment: !reset []
    ports:
      - ${TEMPORAL_UI_PORT:-8081}:8080
    volumes:
      - temporal-lite-data:/data
    depends_on: !override
      temporal-lite-init:
        condition: service_completed_successfully
    healthcheck:
      test: ["CMD", "temporal", "operator", "cluster", "health", "--address", "127.0.0.1:7233"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 120s
      start_interval: 2s
    mem_limit: 512m

  temporal_postgres_db:
    profiles: [lite-disabled]

  temporal_ui:
    profiles: [lite-disabled]
    # The dev server serves the UI on this port; avoid a clash if every
    # profile is enabled (COMPOSE_PROFILES='*').
    ports: !reset []

  litellm:
    profiles: [agents]
    environment: *python-env
    healthcheck:
      interval: 30s
      start_period: 180s
      start_interval: 2s
      retries: 3

  agent-worker:
    profiles: [agents]
    environment: *python-env

  agent-executor:
    profiles: [agents]
    environment: *python-env

  mcp:
    profiles: [mcp]
    environment: *python-env

  background-worker:
    environment: *python-env
    mem_limit: 1g

  api:
    environment: *python-env
    healthcheck:
      interval: 30s
      start_period: 180s
      start_interval: 2s
      retries: 3
    mem_limit: 1536m

  worker:
    environment: *python-env
    mem_limit: 1g

  executor:
    environment: *python-env
    # Above one sandbox's default budget (TRACECAT__SANDBOX_DEFAULT_MEMORY_MB,
    # 2048) plus the executor process, so one action cannot OOM the executor.
    mem_limit: 3g

  migrations:
    environment: *python-env

  ui:
    mem_limit: 3g

  postgres_db:
    command:
      - postgres
      - -c
      - shared_buffers=64MB
    healthcheck:
      interval: 30s
      start_interval: 2s
    mem_limit: 1g

  minio:
    healthcheck:
      interval: 30s
      start_period: 60s
      start_interval: 2s
    mem_limit: 512m

  redis:
    healthcheck:
      interval: 30s
      start_period: 30s
      start_interval: 1s
    mem_limit: 256m

  caddy:
    mem_limit: 128m

volumes:
  temporal-lite-data: