docker-compose.yml

# Unified docker-compose: one file for local dev, CI, and the real-LLM
# harness tier.
#
# The three previous overlays (docker-compose.{ci,real-llm}.yml) are
# replaced by env-var substitution + optional service profiles.  Same
# set of core services (postgres + api + dashboard + inngest-dev) runs
# everywhere; only the secrets/passwords/profiles change per caller.
#
# Usage
# -----
#
# Local dev (defaults — dashboard + api hot-reload, observability off):
#
#     docker compose up -d
#
# CI (ephemeral postgres password; minimal set; BuildKit GHA cache):
#
#     POSTGRES_PASSWORD=ci_test \
#     docker compose up -d postgres api inngest-dev dashboard
#
# Real-LLM harness (provider keys):
#
#     POSTGRES_PASSWORD=ergon \
#     OPENROUTER_API_KEY="$OPENROUTER_API_KEY" \
#     OPENAI_API_KEY="$OPENAI_API_KEY" \
#     EXA_API_KEY="$EXA_API_KEY" \
#     HF_API_KEY="$HF_API_KEY" \
#     docker compose up -d
#
# Observability stack (otel + jaeger) on demand:
#
#     docker compose --profile observability up -d
#
# Service port conventions (host:container):
#   postgres    5433:5432
#   api         9000:9000
#   dashboard   3001:3000
#   inngest-dev 8289:8288
#
# All of the above can be overridden via env vars without editing this
# file (see the ``${VAR:-default}`` forms).

services:
  postgres:
    image: postgres:15
    environment:
      POSTGRES_USER: ergon
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-ergon_dev}
      POSTGRES_DB: ergon
    ports:
      - "5433:5432"
    volumes:
      - postgres_data:/var/lib/postgresql/data
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U ergon"]
      interval: 2s
      timeout: 3s
      retries: 10

  api:
    build:
      context: .
      dockerfile: Dockerfile
      # BuildKit GHA cache: CI passes these through via buildx.  Local
      # developers get a no-op (``type=gha`` is only honoured when the
      # buildx GHA-cache endpoints are present, which GitHub Actions
      # sets automatically).  No local setup required.
      cache_from:
        - type=gha,scope=ergon-api
      cache_to:
        - type=gha,scope=ergon-api,mode=max
    image: ergon-api:local
    ports:
      - "9000:9000"
    env_file:
      - path: .env
        required: false
    environment:
      - ERGON_DATABASE_URL=postgresql://ergon:${POSTGRES_PASSWORD:-ergon_dev}@postgres:5432/ergon
      - INNGEST_DEV=1
      - INNGEST_EVENT_KEY=dev
      - INNGEST_API_BASE_URL=http://inngest-dev:8288
      - ERGON_API_BASE_URL=http://api:9000
      - ERGON_BLOB_ROOT=/tmp/ergon-blob
      - OTEL_TRACES_ENABLED=false
      - OTEL_SERVICE_NAME=ergon-core
      - E2B_API_KEY=${E2B_API_KEY:-}
      - OPENROUTER_API_KEY=${OPENROUTER_API_KEY:-}
      - OPENAI_API_KEY=${OPENAI_API_KEY:-}
      - EXA_API_KEY=${EXA_API_KEY:-}
      - HF_API_KEY=${HF_API_KEY:-}
      # Put /app on sys.path so editable source mounts resolve in the API
      # container while local smoke fixtures live under tests.fixtures.
      - PYTHONPATH=/app
      # Unbuffered Python stdio so ``docker compose logs api`` streams
      # logger output as it happens.  Default (buffered) hides
      # exceptions behind the uvicorn worker's output queue for many
      # seconds, turning any "why is this stuck" investigation into a
      # long stare at stale logs.
      - PYTHONUNBUFFERED=1
    volumes:
      # Source mounts: support ``--reload`` in dev.  Harmless in CI —
      # the same source is on the runner's filesystem from ``checkout``.
      - ./ergon_core:/app/ergon_core
      - ./ergon_builtins:/app/ergon_builtins
      - ./ergon_cli:/app/ergon_cli
      - ./tests:/app/tests
      - ./data:/app/data
      # Blob store bind mount so host-side pytest can read files the
      # SandboxResourcePublisher writes from inside the container.  The
      # mount uses the same path on both sides so the DB-stored
      # ``file_path`` (computed from ``ERGON_BLOB_ROOT``) resolves on
      # both ends.
      - /tmp/ergon-blob:/tmp/ergon-blob
    depends_on:
      postgres:
        condition: service_healthy
    # Without this, ``docker compose up --wait`` only waits for Postgres.
    # Uvicorn + ``--reload`` can still be importing when CI starts Playwright,
    # so ``GET /api/health`` in the dashboard sees a failing ``ergon_api``
    # probe and returns 503 (see ``ergon-dashboard/tests/e2e/health.spec.ts``).
    healthcheck:
      test: ["CMD-SHELL", "curl -fsS http://127.0.0.1:9000/health >/dev/null || exit 1"]
      interval: 3s
      timeout: 5s
      retries: 30
      start_period: 120s
    command: >
      uvicorn ergon_cli.api_app:app
      --host 0.0.0.0 --port 9000 --reload
      --reload-dir /app/ergon_core
      --reload-dir /app/ergon_builtins
      --reload-dir /app/tests
      --timeout-graceful-shutdown 30

  dashboard:
    build:
      context: ./ergon-dashboard
      dockerfile: Dockerfile
    ports:
      - "3001:3000"
    environment:
      - NODE_ENV=development
      - ERGON_API_BASE_URL=http://api:9000
      - INNGEST_DEV=1
      - INNGEST_EVENT_KEY=dev
      - INNGEST_API_BASE_URL=http://inngest-dev:8288
      - WATCHPACK_POLLING=true
    depends_on:
      - inngest-dev
      - api
    command: pnpm dev
    volumes:
      - ./ergon-dashboard:/app
      - /app/node_modules
      - /app/.next

  inngest-dev:
    image: inngest/inngest:latest
    command: >
      inngest dev --no-discovery
      -u http://api:9000/api/inngest
      -u http://dashboard:3000/api/inngest
    ports:
      - "8289:8288"
    environment:
      - INNGEST_DEV=1
    depends_on:
      api:
        condition: service_started

  # -- Observability stack (opt-in via ``--profile observability``) ---
  #
  # Not started by default.  Enables OTel trace export to Jaeger for
  # debugging runtime + execution ordering.
  otel-collector:
    profiles: ["observability"]
    image: otel/opentelemetry-collector-contrib:latest
    command: ["--config=/etc/otelcol-contrib/config.yaml"]
    volumes:
      - ./ergon_infra/ergon_infra/templates/otel-collector.yaml:/etc/otelcol-contrib/config.yaml
    ports:
      - "4319:4317"

  jaeger:
    profiles: ["observability"]
    image: jaegertracing/all-in-one:latest
    ports:
      - "16687:16686"
      - "4320:4318"
    environment:
      - COLLECTOR_OTLP_ENABLED=true

volumes:
  postgres_data: