docker-compose.yml

# =============================================================================
# TelemetryFlow Python SDK - Docker Compose
# =============================================================================
#
# TelemetryFlow Python SDK - Community Enterprise Observability Platform (CEOP)
# Copyright (c) 2024-2026 Telemetri Data Indonesia. All rights reserved.
#
# PROFILES:
# - dev:       Development environment (tfo-collector)
# - full:      Full observability stack (+ Prometheus)
# - platform:  Full TFO Platform stack (+ tfo-backend, tfo-viz, postgres, clickhouse, redis, nats)
# - example:   SDK example runners
#
# Usage:
#   # Build SDK image
#   docker-compose build
#
#   # Run SDK generator (init command)
#   docker-compose run --rm sdk-gen init --project myapp --service my-service --output /workspace
#
#   # Run RESTful API generator
#   docker-compose run --rm sdk-restapi new --name my-api
#
#   # Start development environment with observability stack
#   docker-compose --profile dev up -d tfo-collector
#
#   # Start full observability stack
#   docker-compose --profile full up -d
#
#   # Start full TFO Platform stack (backend + viz + infrastructure)
#   docker-compose --profile platform up -d
#
# =============================================================================
# OTLP HTTP Endpoints (Dual Ingestion Support)
# =============================================================================
#
#   TelemetryFlow Platform (Recommended - TFO Standalone):
#     POST http://localhost:4318/v2/traces
#     POST http://localhost:4318/v2/metrics
#     POST http://localhost:4318/v2/logs
#
#   OTEL Community (Backwards Compatible - OCB/Standard):
#     POST http://localhost:4318/v1/traces
#     POST http://localhost:4318/v1/metrics
#     POST http://localhost:4318/v1/logs
#
#   gRPC (Both versions via same port):
#     localhost:4317
#
#   NOTE: v2 endpoints require TFO Standalone build.
#         OCB builds support v1 endpoints only.
#
# =============================================================================

#================================================================================================
# NETWORK SETUP
#================================================================================================
networks:
  telemetryflow_python_sdk_net:
    name: telemetryflow_python_sdk_net
    driver: bridge
    ipam:
      config:
        - subnet: 172.153.0.0/16

#================================================================================================
# SERVICES
#================================================================================================
services:

  #==============================================================================================
  # SDK SERVICES
  #==============================================================================================

  #----------------------------------------------------------------------------------------------
  # SDK Generator
  #----------------------------------------------------------------------------------------------
  sdk-gen:
    container_name: ${CONTAINER_NAME:-telemetryflow_sdk_gen}
    image: ${IMAGE_NAME:-telemetryflow/telemetryflow-python-sdk}:${IMAGE_TAG:-1.2.0}
    build:
      context: .
      dockerfile: Dockerfile
      args:
        VERSION: ${VERSION:-1.2.0}
        GIT_COMMIT: ${GIT_COMMIT:-unknown}
        GIT_BRANCH: ${GIT_BRANCH:-main}
        BUILD_TIME: ${BUILD_TIME:-unknown}
    entrypoint: ["telemetryflow-gen"]
    command: ["--help"]
    volumes:
      - ./workspace:/workspace
    working_dir: /workspace
    networks:
      - telemetryflow_python_sdk_net

  #----------------------------------------------------------------------------------------------
  # RESTful API Generator
  #----------------------------------------------------------------------------------------------
  sdk-restapi:
    container_name: ${CONTAINER_NAME_RESTAPI:-telemetryflow_sdk_restapi}
    image: ${IMAGE_NAME:-telemetryflow/telemetryflow-python-sdk}:${IMAGE_TAG:-1.2.0}
    build:
      context: .
      dockerfile: Dockerfile
      args:
        VERSION: ${VERSION:-1.2.0}
        GIT_COMMIT: ${GIT_COMMIT:-unknown}
        GIT_BRANCH: ${GIT_BRANCH:-main}
        BUILD_TIME: ${BUILD_TIME:-unknown}
    entrypoint: ["python"]
    command: ["-c", "from telemetryflow import TelemetryFlowBuilder; print('SDK Ready!')"]
    environment:
      - TELEMETRYFLOW_API_KEY_ID=${TELEMETRYFLOW_API_KEY_ID:-tfk_example}
      - TELEMETRYFLOW_API_KEY_SECRET=${TELEMETRYFLOW_API_KEY_SECRET:-tfs_example}
      - TELEMETRYFLOW_ENDPOINT=${TELEMETRYFLOW_ENDPOINT:-tfo-collector:4317}
      - TELEMETRYFLOW_SERVICE_NAME=${TELEMETRYFLOW_SERVICE_NAME:-telemetryflow-python-sdk}
      - TELEMETRYFLOW_SERVICE_VERSION=${TELEMETRYFLOW_SERVICE_VERSION:-1.2.0}
      - TELEMETRYFLOW_ENVIRONMENT=${TELEMETRYFLOW_ENVIRONMENT:-development}
    volumes:
      - ./workspace:/workspace
    working_dir: /workspace
    depends_on:
      tfo-collector:
        condition: service_healthy
    networks:
      telemetryflow_python_sdk_net:
        ipv4_address: ${CONTAINER_IP_SDK_RESTAPI:-172.153.152.10}
    profiles:
      - example

  #----------------------------------------------------------------------------------------------
  # SDK Development Container
  #----------------------------------------------------------------------------------------------
  sdk-dev:
    container_name: ${CONTAINER_NAME_DEV:-telemetryflow_sdk_dev}
    image: ${IMAGE_NAME:-telemetryflow/telemetryflow-python-sdk}:${IMAGE_TAG:-1.2.0}
    build:
      context: .
      dockerfile: Dockerfile.dev
      args:
        VERSION: ${VERSION:-1.2.0}
        GIT_COMMIT: ${GIT_COMMIT:-unknown}
        GIT_BRANCH: ${GIT_BRANCH:-main}
        BUILD_TIME: ${BUILD_TIME:-unknown}
    entrypoint: ["python"]
    command: ["-c", "from telemetryflow import TelemetryFlowBuilder; print('SDK Ready!')"]
    environment:
      - TELEMETRYFLOW_API_KEY_ID=${TELEMETRYFLOW_API_KEY_ID:-tfk_example}
      - TELEMETRYFLOW_API_KEY_SECRET=${TELEMETRYFLOW_API_KEY_SECRET:-tfs_example}
      - TELEMETRYFLOW_ENDPOINT=${TELEMETRYFLOW_ENDPOINT:-tfo-collector:4317}
      - TELEMETRYFLOW_SERVICE_NAME=${TELEMETRYFLOW_SERVICE_NAME:-telemetryflow-python-sdk-dev}
      - TELEMETRYFLOW_SERVICE_VERSION=${TELEMETRYFLOW_SERVICE_VERSION:-1.2.0}
      - TELEMETRYFLOW_ENVIRONMENT=${TELEMETRYFLOW_ENVIRONMENT:-development}
    volumes:
      - .:/app
      - ./workspace:/workspace
    working_dir: /app
    depends_on:
      tfo-collector:
        condition: service_healthy
    networks:
      telemetryflow_python_sdk_net:
        ipv4_address: ${CONTAINER_IP_SDK_DEV:-172.153.152.11}
    profiles:
      - dev

  #==============================================================================================
  # MONITORING SERVICES (aligned with TFO-Platform)
  #==============================================================================================

  #----------------------------------------------------------------------------------------------
  # TFO-COLLECTOR - TelemetryFlow Collector (OCB Native v1.2.1)
  #----------------------------------------------------------------------------------------------
  # Supports dual ingestion: v1 (OTEL standard) and v2 (TelemetryFlow enhanced)
  #
  # OTLP HTTP Endpoints:
  #   TelemetryFlow Platform (Recommended - TFO Standalone):
  #     POST http://localhost:4318/v2/traces
  #     POST http://localhost:4318/v2/metrics
  #     POST http://localhost:4318/v2/logs
  #
  #   OTEL Community (Backwards Compatible - OCB/Standard):
  #     POST http://localhost:4318/v1/traces
  #     POST http://localhost:4318/v1/metrics
  #     POST http://localhost:4318/v1/logs
  #
  #   gRPC (Both versions via same port): localhost:4317
  #----------------------------------------------------------------------------------------------
  tfo-collector:
    profiles: ["dev", "full", "example"]
    platform: linux/amd64
    # =============================================================================
    # TelemetryFlow Collector (TFO-Collector) - Custom TFO format (v1.2.1)
    image: telemetryflow/telemetryflow-collector:${TFO_COLLECTOR_VERSION:-1.2.1}
    command: ["--config=/etc/tfo-collector/tfo-collector.yaml"]
    # =============================================================================
    # OTEL Collector Community Contributor (Standard OTEL format - uncomment to use)
    # image: otel/opentelemetry-collector-contrib:${OTEL_VERSION:-0.142.0}
    # command: ["--config=/etc/otelcol-contrib/config.yaml"]
    # =============================================================================
    container_name: ${CONTAINER_OTEL:-telemetryflow_sdk_tfo_collector}
    restart: unless-stopped
    environment:
      # Backend endpoint for tfo exporter (host.docker.internal → host)
      - TELEMETRYFLOW_ENDPOINT=${TELEMETRYFLOW_BACKEND_ENDPOINT:-http://host.docker.internal:3000}
      # API keys for tfoauth extension
      - TELEMETRYFLOW_API_KEY_ID=${TELEMETRYFLOW_API_KEY_ID}
      - TELEMETRYFLOW_API_KEY_SECRET=${TELEMETRYFLOW_API_KEY_SECRET}
      # Collector identity
      - TELEMETRYFLOW_COLLECTOR_ID=${CONTAINER_OTEL:-telemetryflow_sdk_tfo_collector}
      - TELEMETRYFLOW_ENVIRONMENT=${TELEMETRYFLOW_ENVIRONMENT:-development}
    volumes:
      # TelemetryFlow Collector config (Custom TFO format)
      - ./configs/tfo-collector.yaml:/etc/tfo-collector/tfo-collector.yaml:ro
      # =============================================================================
      # OTEL Community Collector config (uncomment if using otel/opentelemetry-collector-contrib image)
      # - ./configs/otel-collector.yaml:/etc/otelcol-contrib/config.yaml:ro
    ports:
      - "${PORT_OTEL_GRPC:-4317}:4317"     # OTLP gRPC (v1 & v2)
      - "${PORT_OTEL_HTTP:-4318}:4318"     # OTLP HTTP (v1 & v2)
      - "${PORT_OTEL_METRICS:-8889}:8889"  # Prometheus exporter
      - "${PORT_OTEL_HEALTH:-13133}:13133" # Health check
      - "${PORT_OTEL_ZPAGES:-55679}:55679" # zPages
      - "${PORT_OTEL_PPROF:-1777}:1777"    # pprof
    extra_hosts:
      - "host.docker.internal:host-gateway"
    healthcheck:
      test: ["CMD", "wget", "--spider", "-q", "http://localhost:13133"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 10s
    networks:
      telemetryflow_python_sdk_net:
        ipv4_address: ${CONTAINER_IP_OTEL:-172.153.152.30}

  #----------------------------------------------------------------------------------------------
  # PROMETHEUS (for metrics visualization)
  #----------------------------------------------------------------------------------------------
  prometheus:
    container_name: ${CONTAINER_PROMETHEUS:-telemetryflow_sdk_prometheus}
    image: prom/prometheus:${PROMETHEUS_VERSION:-v3.4.0}
    ports:
      - "${PORT_PROMETHEUS:-9090}:9090"
    volumes:
      - ./configs/prometheus.yaml:/etc/prometheus/prometheus.yml:ro
      - prometheus-data:/prometheus
    command:
      - '--config.file=/etc/prometheus/prometheus.yml'
      - '--storage.tsdb.path=/prometheus'
      - '--web.enable-lifecycle'
    networks:
      telemetryflow_python_sdk_net:
        ipv4_address: ${CONTAINER_IP_PROMETHEUS:-172.153.152.70}
    profiles:
      - full

  #==============================================================================================
  # PLATFORM SERVICES (aligned with TFO-Platform monolith)
  #==============================================================================================

  #----------------------------------------------------------------------------------------------
  # POSTGRESQL - Relational Data Storage (Config, Entities, IAM)
  #----------------------------------------------------------------------------------------------
  postgres:
    profiles: ["platform"]
    platform: linux/amd64
    image: postgres:${POSTGRES_VERSION:-16-alpine}
    container_name: ${CONTAINER_POSTGRES:-telemetryflow_sdk_postgres}
    restart: unless-stopped
    ports:
      - "${PORT_POSTGRES:-5432}:5432"
    environment:
      - TZ=${TZ:-UTC}
      - POSTGRES_USER=${POSTGRES_USERNAME:-tfo_admin}
      - POSTGRES_PASSWORD=${POSTGRES_PASSWORD:?POSTGRES_PASSWORD must be set}
      - POSTGRES_DB=${POSTGRES_DB:-telemetryflow_db}
      - PGDATA=/var/lib/postgresql/data/pgdata
    volumes:
      - ${DATA_POSTGRESQL:-${VOLUMES_BASE_PATH:-/opt/data/docker/telemetryflow-sdk}/postgresql}:/var/lib/postgresql/data
    networks:
      telemetryflow_python_sdk_net:
        ipv4_address: ${CONTAINER_IP_POSTGRES:-172.153.153.20}
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U ${POSTGRES_USERNAME:-tfo_admin} -d ${POSTGRES_DB:-telemetryflow_db}"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 10s

  #----------------------------------------------------------------------------------------------
  # CLICKHOUSE - Time-Series & Analytics Storage (Metrics, Logs, Traces, Audit)
  #----------------------------------------------------------------------------------------------
  clickhouse:
    profiles: ["platform"]
    platform: linux/amd64
    image: clickhouse/clickhouse-server:${CLICKHOUSE_VERSION:-latest}
    container_name: ${CONTAINER_CLICKHOUSE:-telemetryflow_sdk_clickhouse}
    restart: unless-stopped
    ports:
      - "${PORT_CLICKHOUSE_HTTP:-8123}:8123"    # HTTP
      - "${PORT_CLICKHOUSE_NATIVE:-9000}:9000"  # Native
    environment:
      - TZ=${TZ:-UTC}
      - CLICKHOUSE_DB=${CLICKHOUSE_DB:-telemetryflow_db}
      - CLICKHOUSE_USER=${CLICKHOUSE_USER:-tfo_admin}
      - CLICKHOUSE_PASSWORD=${CLICKHOUSE_PASSWORD:?CLICKHOUSE_PASSWORD must be set}
    volumes:
      - ${DATA_CLICKHOUSE:-${VOLUMES_BASE_PATH:-/opt/data/docker/telemetryflow-sdk}/clickhouse}/data:/var/lib/clickhouse
      - ${DATA_CLICKHOUSE:-${VOLUMES_BASE_PATH:-/opt/data/docker/telemetryflow-sdk}/clickhouse}/logs:/var/log/clickhouse-server
    networks:
      telemetryflow_python_sdk_net:
        ipv4_address: ${CONTAINER_IP_CLICKHOUSE:-172.153.153.40}
    healthcheck:
      test: ["CMD", "wget", "--spider", "-q", "http://localhost:8123/ping"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 10s

  #----------------------------------------------------------------------------------------------
  # REDIS - Cache (L2), Session Store, BullMQ Job Queues
  #----------------------------------------------------------------------------------------------
  redis:
    profiles: ["platform"]
    platform: linux/amd64
    image: redis:${REDIS_VERSION:-7-alpine}
    container_name: ${CONTAINER_REDIS:-telemetryflow_sdk_redis}
    restart: unless-stopped
    ports:
      - "${PORT_REDIS:-6379}:6379"
    command: >
      redis-server
      --maxmemory ${REDIS_MAX_MEMORY:-512mb}
      --maxmemory-policy allkeys-lru
      --appendonly yes
      --appendfsync everysec
    volumes:
      - ${DATA_REDIS:-${VOLUMES_BASE_PATH:-/opt/data/docker/telemetryflow-sdk}/redis}:/data
    networks:
      telemetryflow_python_sdk_net:
        ipv4_address: ${CONTAINER_IP_REDIS:-172.153.153.50}
    healthcheck:
      test: ["CMD", "redis-cli", "ping"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 5s

  #----------------------------------------------------------------------------------------------
  # NATS - Cross-Module Domain Event Messaging
  #----------------------------------------------------------------------------------------------
  nats:
    profiles: ["platform"]
    platform: linux/amd64
    image: nats:${NATS_VERSION:-2-alpine}
    container_name: ${CONTAINER_NATS:-telemetryflow_sdk_nats}
    restart: unless-stopped
    ports:
      - "${PORT_NATS_CLIENT:-4222}:4222"   # Client connections
      - "${PORT_NATS_MONITOR:-8222}:8222"  # HTTP monitoring
    command: ["--js", "--sd", "/data", "-m", "8222"]
    volumes:
      - ${DATA_NATS:-${VOLUMES_BASE_PATH:-/opt/data/docker/telemetryflow-sdk}/nats}:/data
    networks:
      telemetryflow_python_sdk_net:
        ipv4_address: ${CONTAINER_IP_NATS:-172.153.153.55}
    healthcheck:
      test: ["CMD", "wget", "--spider", "-q", "http://localhost:8222/healthz"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 5s

  #----------------------------------------------------------------------------------------------
  # TFO-BACKEND - NestJS API
  #   Receives telemetry from TFO-Collector via tfo exporter, stores in ClickHouse
  #   Provides REST API for TFO-Viz frontend
  #----------------------------------------------------------------------------------------------
  tfo-backend:
    profiles: ["platform"]
    platform: linux/amd64
    image: telemetryflow/telemetryflow-platform:${TFO_PLATFORM_VERSION:-1.4.0}
    container_name: ${CONTAINER_BACKEND:-telemetryflow_sdk_backend}
    restart: unless-stopped
    ports:
      - "${PORT_BACKEND:-3000}:3000"
    environment:
      - NODE_ENV=${NODE_ENV:-development}
      - PORT=3000
      - TZ=${TZ:-UTC}

      # PostgreSQL
      - POSTGRES_HOST=postgres
      - POSTGRES_PORT=5432
      - POSTGRES_DB=${POSTGRES_DB:-telemetryflow_db}
      - POSTGRES_USERNAME=${POSTGRES_USERNAME:-tfo_admin}
      - POSTGRES_PASSWORD=${POSTGRES_PASSWORD:?POSTGRES_PASSWORD must be set}

      # ClickHouse
      - CLICKHOUSE_HOST=clickhouse
      - CLICKHOUSE_PORT=8123
      - CLICKHOUSE_DB=${CLICKHOUSE_DB:-telemetryflow_db}
      - CLICKHOUSE_USER=${CLICKHOUSE_USER:-tfo_admin}
      - CLICKHOUSE_PASSWORD=${CLICKHOUSE_PASSWORD:?CLICKHOUSE_PASSWORD must be set}

      # Redis
      - REDIS_HOST=redis
      - REDIS_PORT=6379
      - REDIS_PASSWORD=${REDIS_PASSWORD:-}
      - REDIS_MAX_MEMORY=${REDIS_MAX_MEMORY:-512mb}
      - REDIS_DB_SESSION=0
      - REDIS_DB_CACHE=1
      - REDIS_DB_BULLMQ=2

      # NATS
      - NATS_URL=nats://nats:4222

      # BullMQ
      - ENABLE_BULLMQ=${ENABLE_BULLMQ:-true}

      # JWT & Session
      - JWT_SECRET=${JWT_SECRET:?JWT_SECRET must be set}
      - JWT_EXPIRES_IN=${JWT_EXPIRES_IN:-24h}
      - SESSION_SECRET=${SESSION_SECRET:?SESSION_SECRET must be set}

      # CORS
      - CORS_ORIGIN=${CORS_ORIGIN:-http://localhost:8080}

      # Logging
      - LOGGER_TYPE=${LOGGER_TYPE:-winston}
      - LOG_LEVEL=${LOG_LEVEL:-info}
      - LOG_PRETTY_PRINT=${LOG_PRETTY_PRINT:-false}

      # OpenTelemetry (disabled — SDK dev uses collector directly)
      - OTEL_ENABLED=false
      - OTEL_SERVICE_NAME=${OTEL_SERVICE_NAME:-telemetryflow-platform}
    depends_on:
      postgres:
        condition: service_healthy
      clickhouse:
        condition: service_healthy
      redis:
        condition: service_healthy
      nats:
        condition: service_healthy
    networks:
      telemetryflow_python_sdk_net:
        ipv4_address: ${CONTAINER_IP_BACKEND:-172.153.153.10}
    healthcheck:
      test: ["CMD-SHELL", "wget --no-verbose --tries=1 --spider http://localhost:3000/health || exit 1"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 40s

  #----------------------------------------------------------------------------------------------
  # TFO-VIZ - Vue 3 SPA Frontend (Nginx)
  #   Telemetry visualization dashboard — proxies /api to tfo-backend
  #----------------------------------------------------------------------------------------------
  tfo-viz:
    profiles: ["platform"]
    platform: linux/amd64
    image: telemetryflow/telemetryflow-viz:${TFO_PLATFORM_VERSION:-1.4.0}
    container_name: ${CONTAINER_FRONTEND:-telemetryflow_sdk_frontend}
    restart: unless-stopped
    ports:
      - "${PORT_FRONTEND:-8080}:80"
    depends_on:
      tfo-backend:
        condition: service_healthy
    networks:
      telemetryflow_python_sdk_net:
        ipv4_address: ${CONTAINER_IP_FRONTEND:-172.153.153.15}
    healthcheck:
      test: ["CMD", "wget", "--spider", "-q", "http://localhost:80/health"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 10s

#================================================================================================
# VOLUMES
#================================================================================================
volumes:
  prometheus-data:
    driver: local