docker-compose.yaml

include:
  - path:
    - nims.yaml

services:
  # =======================
  # Agent Services
  # =======================
  agent-chain-server:
    container_name: agent-chain-server
    image: nvcr.io/nvidia/blueprint/aiva-customer-service-agent:1.1.0
    build:
      # Set context to repo's root directory
      context: ../../
      dockerfile: src/agent/Dockerfile
    command: --port 8081 --host 0.0.0.0 --workers 1 --loop asyncio
    environment:
      EXAMPLE_PATH: './src/agent'
      #APP_LLM_MODELNAME: ${APP_LLM_MODELNAME:-"meta/llama-3.1-70b-instruct"}
      APP_LLM_MODELNAME: ${APP_LLM_MODELNAME:-"meta/llama-3.1-405b-instruct"}
      APP_LLM_MODELENGINE: nvidia-ai-endpoints
      APP_LLM_SERVERURL: ${APP_LLM_SERVERURL:-""}
      # Cache name to store user conversation
      # supported type inmemory, redis
      APP_CACHE_NAME: ${APP_CACHE_NAME:-"redis"}
      APP_CACHE_URL: ${APP_CACHE_URL:-"redis:6379"}
      # Database name to store user conversation
      # supported type postgres
      APP_DATABASE_NAME: ${APP_DATABASE_NAME:-"postgres"}
      APP_DATABASE_URL: ${APP_DATABASE_URL:-"postgres:5432"}
      # Checkpointer name to store intermediate state of conversation
      # supported type postgres, inmemory
      APP_CHECKPOINTER_NAME: ${APP_CHECKPOINTER_NAME:-"postgres"}
      APP_CHECKPOINTER_URL: ${APP_CHECKPOINTER_URL:-"postgres:5432"}
      # Postgres config
      POSTGRES_USER: ${POSTGRES_USER:-postgres}
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-password}
      POSTGRES_DB: ${POSTGRES_DB:-postgres}
      # Postgres database name for customer data
      POSTGRES_USER_READONLY: ${POSTGRES_USER:-postgres_readonly}
      POSTGRES_PASSWORD_READONLY: ${POSTGRES_PASSWORD:-readonly_password}
      CUSTOMER_DATA_DB: ${CUSTOMER_DATA_DB:-customer_data}
      CANONICAL_RAG_URL: http://unstructured-retriever:8081
      STRUCTURED_RAG_URI: http://structured-retriever:8081
      NVIDIA_API_KEY: ${NVIDIA_API_KEY}
      GRAPH_RECURSION_LIMIT: 20
      #GRAPH_TIMEOUT_IN_SEC: 20 # with meta/llama-3.1-70b-instruct
      GRAPH_TIMEOUT_IN_SEC: 50 # with meta/llama-3.1-405b-instruct
      RETURN_WINDOW_CURRENT_DATE: '2024-10-23' # Leave it empty to get the current date
      RETURN_WINDOW_THRESHOLD_DAYS: 30
      # Log level for server, supported level NOTSET, DEBUG, INFO, WARN, ERROR, CRITICAL
      LOGLEVEL: ${LOGLEVEL:-INFO}
      # Set the default expiration time (TTL) for Redis keys (in seconds)
      REDIS_SESSION_EXPIRY: 12 # in hours
    ports:
    - "8081:8081"
    expose:
    - "8081"
    shm_size: 5gb
    depends_on:
    - unstructured-retriever
    - structured-retriever
    - postgres
    - redis


  # =======================
  # UI and related services for Agent Interface
  # =======================
  api-gateway-server:
    build:
      context: ../../
      dockerfile: ./src/api_gateway/Dockerfile
    image: nvcr.io/nvidia/blueprint/aiva-customer-service-api-gateway:1.1.0
    command: --port 9000 --host 0.0.0.0 --workers 1
    ports:
      - "9000:9000"
    environment:
      AGENT_SERVER_URL: ${AGENT_SERVER_URL:-http://agent-chain-server:8081}
      ANALYTICS_SERVER_URL: ${ANALYTICS_SERVER_URL:-http://analytics-server:8081}
      #REQUEST_TIMEOUT: 320 # with meta/llama-3.1-70b-instruct
      REQUEST_TIMEOUT: 800  # with meta/llama-3.1-405b-instruct
    restart: unless-stopped  # Optional: Automatically restart the container unless it is stopped
    depends_on:
    - agent-chain-server

  agent-frontend:
    image: nvcr.io/nvidia/blueprint/aiva-customer-service-ui:1.1.0
    container_name: agent-frontend
    environment:
      - INFERENCE_ORIGIN=http://api-gateway-server:9000
    ports:
      - "3001:3001"
    restart: unless-stopped  # Optional: Automatically restart the container unless it is stopped
    depends_on:
    - api-gateway-server

  # =======================
  # Analytics Services - summary/sentiment and similar APIs are exposed as part of analytics MS
  # =======================
  analytics-server:
    container_name: analytics-server
    image: nvcr.io/nvidia/blueprint/aiva-customer-service-analytics:1.1.0
    build:
      # Set context to repo's root directory
      context: ../../
      dockerfile: src/analytics/Dockerfile
    command: --port 8081 --host 0.0.0.0 --workers 1
    environment:
      EXAMPLE_PATH: './src/analytics'
      #APP_LLM_MODELNAME: ${APP_LLM_MODELNAME:-"meta/llama-3.1-70b-instruct"}
      APP_LLM_MODELNAME: ${APP_LLM_MODELNAME:-"meta/llama-3.1-405b-instruct"}
      APP_LLM_MODELENGINE: nvidia-ai-endpoints
      APP_LLM_SERVERURL: ${APP_LLM_SERVERURL:-""}
      # Database name to store user conversation/summary
      # supported type inmemory, redis
      APP_DATABASE_NAME: ${APP_DATABASE_NAME:-"postgres"}
      APP_DATABASE_URL: ${APP_DATABASE_URL:-"postgres:5432"}
      # Postgres config
      POSTGRES_USER: ${POSTGRES_USER:-postgres}
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-password}
      POSTGRES_DB: ${POSTGRES_DB:-postgres}
      # Postgres database name for customer data
      CUSTOMER_DATA_DB: ${CUSTOMER_DATA_DB:-customer_data}
      # Store summary/sentiment in database
      PERSIST_DATA: ${PERSIST_DATA:-true}
      NVIDIA_API_KEY: ${NVIDIA_API_KEY}
      # Log level for server, supported level NOTSET, DEBUG, INFO, WARN, ERROR, CRITICAL
      LOGLEVEL: ${LOGLEVEL:-INFO}
    ports:
    - "8082:8081"
    expose:
    - "8081"
    shm_size: 5gb
    depends_on:
      postgres:
        condition: service_healthy


  # =======================
  # Retriever Microservices
  # =======================
  # Fetch relevant document from vectorstore
  unstructured-retriever:
    container_name: unstructured-retriever
    image: nvcr.io/nvidia/blueprint/aiva-customer-service-unstructured-retriever:1.1.0
    build:
      # Set context to repo's root directory
      context: ../../
      dockerfile: src/retrievers/Dockerfile
      args:
        # Build args, used to copy relevant directory inside the container
        EXAMPLE_PATH: 'src/retrievers/unstructured_data'
    # start the server on port 8081
    command: --port 8081 --host 0.0.0.0 --workers 1
    environment:
      # Path to example directory relative to GenerativeAIExamples/RAG/examples
      EXAMPLE_PATH: 'src/retrievers/unstructured_data'
      # URL on which vectorstore is hosted
      APP_VECTORSTORE_URL: "http://milvus:19530"
      # Type of vectordb used to store embedding supported type milvus, pgvector
      APP_VECTORSTORE_NAME: "milvus"
      # url on which llm model is hosted. If "", Nvidia hosted API is used
      #APP_LLM_MODELNAME: ${APP_LLM_MODELNAME:-"meta/llama-3.1-70b-instruct"}
      APP_LLM_MODELNAME: ${APP_LLM_MODELNAME:-"meta/llama-3.1-405b-instruct"}
      # embedding model engine used for inference, supported type nvidia-ai-endpoints, huggingface
      APP_LLM_MODELENGINE: nvidia-ai-endpoints
      # url on which llm model is hosted. If "", Nvidia hosted API is used
      APP_LLM_SERVERURL: ${APP_LLM_SERVERURL:-""}
      APP_EMBEDDINGS_MODELNAME: ${APP_EMBEDDINGS_MODELNAME:-nvidia/nv-embedqa-e5-v5}
      # embedding model engine used for inference, supported type nvidia-ai-endpoints
      APP_EMBEDDINGS_MODELENGINE: ${APP_EMBEDDINGS_MODELENGINE:-nvidia-ai-endpoints}
      # url on which embedding model is hosted. If "", Nvidia hosted API is used
      APP_EMBEDDINGS_SERVERURL: ${APP_EMBEDDINGS_SERVERURL:-""}
      APP_RANKING_MODELNAME: ${APP_RANKING_MODELNAME:-"nvidia/nv-rerankqa-mistral-4b-v3"} # Leave it blank to avoid using ranking
      # ranking engine used for inference, supported type nvidia-ai-endpoints
      APP_RANKING_MODELENGINE: ${APP_RANKING_MODELENGINE:-nvidia-ai-endpoints}
      # url on which re-ranking model is hosted. If "", Nvidia hosted API is used
      APP_RANKING_SERVERURL: ${APP_RANKING_SERVERURL:-""}
      # text splitter model name, it's fetched from huggingface
      APP_TEXTSPLITTER_MODELNAME: Snowflake/snowflake-arctic-embed-l
      APP_TEXTSPLITTER_CHUNKSIZE: 506
      APP_TEXTSPLITTER_CHUNKOVERLAP: 200
      NVIDIA_API_KEY: ${NVIDIA_API_KEY}
      # vectorstore collection name to store embeddings
      COLLECTION_NAME: ${COLLECTION_NAME:-unstructured_data}
      APP_RETRIEVER_TOPK: 4
      APP_RETRIEVER_SCORETHRESHOLD: 0.25
      # Number of documents to be retrieved from retriever when reranking model is enabled
      # This will be then send to re ranker to get `APP_RETRIEVER_TOPK` documents
      VECTOR_DB_TOPK: 20
      # Log level for server, supported level NOTSET, DEBUG, INFO, WARN, ERROR, CRITICAL
      LOGLEVEL: ${LOGLEVEL:-INFO}
    ports:
    - "8086:8081"
    expose:
    - "8081"
    shm_size: 5gb
    depends_on:
    - milvus


  # Fetch user information form database
  structured-retriever:
    container_name: structured-retriever
    image: nvcr.io/nvidia/blueprint/aiva-customer-service-structured-retriever:1.1.0
    build:
      context: ../../
      dockerfile: src/retrievers/Dockerfile
      args:
        EXAMPLE_PATH: 'src/retrievers/structured_data'
    command: --port 8081 --host 0.0.0.0 --workers 1
    environment:
      EXAMPLE_PATH: 'src/retrievers/structured_data'
      #APP_LLM_MODELNAME: ${APP_LLM_MODELNAME:-meta/llama-3.1-70b-instruct}
      APP_LLM_MODELNAME: ${APP_LLM_MODELNAME:-meta/llama-3.1-405b-instruct}
      APP_LLM_MODELENGINE: nvidia-ai-endpoints
      APP_LLM_SERVERURL: ${APP_LLM_SERVERURL:-""}
      APP_EMBEDDINGS_MODELNAME: ${APP_EMBEDDINGS_MODELNAME:-nvidia/nv-embedqa-e5-v5}
      APP_EMBEDDINGS_MODELENGINE: ${APP_EMBEDDINGS_MODELENGINE:-nvidia-ai-endpoints}
      APP_EMBEDDINGS_SERVERURL: ${APP_EMBEDDINGS_SERVERURL:-""}
      APP_PROMPTS_CHATTEMPLATE: "You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Please ensure that your responses are positive in nature."
      APP_PROMPTS_RAGTEMPLATE: "You are a helpful AI assistant named Envie. You will reply to questions only based on the context that you are provided. If something is out of context, you will refrain from replying and politely decline to respond to the user."
      NVIDIA_API_KEY: ${NVIDIA_API_KEY}
      COLLECTION_NAME: ${COLLECTION_NAME:-structured_data}
      APP_VECTORSTORE_URL: "http://milvus:19530"
      APP_VECTORSTORE_NAME: "milvus"
      # Database name to store user purchase history, only postgres is supported
      APP_DATABASE_NAME: ${APP_DATABASE_NAME:-"postgres"}
      APP_DATABASE_URL: ${APP_DATABASE_URL:-"postgres:5432"}
      POSTGRES_USER: ${POSTGRES_USER:-postgres_readonly}
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-readonly_password}
      POSTGRES_DB: ${POSTGRES_DB:-customer_data}
      CSV_NAME: PdM_machines
      LOGLEVEL: ${LOGLEVEL:-INFO}
    ports:
    - "8087:8081"
    expose:
    - "8081"
    shm_size: 5gb
    depends_on:
      postgres:
            condition: service_healthy
            required: false
      milvus:
        condition: service_healthy
      nemollm-embedding:
        condition: service_healthy
        required: false

  # =======================
  # Database Services - User purchase history and permanently store conversation details
  # =======================
  postgres:
    container_name: postgres_container
    image: postgres:17.1
    environment:
      POSTGRES_USER: ${POSTGRES_USER:-postgres}
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-password}
      POSTGRES_DB: ${POSTGRES_DB:-customer_data}
    command:
      - "postgres"
      - "-c"
      - "shared_buffers=256MB"
      - "-c"
      - "max_connections=200"
    volumes:
       - ./init-scripts:/docker-entrypoint-initdb.d  # Mount initialization scripts
       - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/postgres_data:/var/lib/postgresql/data
    ports:
      - "5432:5432"
    healthcheck:
      test: ["CMD-SHELL", "sh -c 'pg_isready -U postgres -d postgres'"]
      interval: 10s
      timeout: 3s
      retries: 3
    restart: unless-stopped

  # For visualization purpose
  pgadmin:
    container_name: pgadmin_container
    image: dpage/pgadmin4:8.13.0
    environment:
      PGADMIN_DEFAULT_EMAIL: ${PGADMIN_DEFAULT_EMAIL:-pgadmin4@pgadmin.org}
      PGADMIN_DEFAULT_PASSWORD: ${PGADMIN_DEFAULT_PASSWORD:-admin}
      PGADMIN_CONFIG_SERVER_MODE: 'False'
    user: '$UID:$GID'
    volumes:
       - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/pgadmin:/var/lib/pgadmin
    ports:
      - "${PGADMIN_PORT:-5050}:80"
    restart: unless-stopped


  # =======================
  # Cache Services - To store conversation of user to share among multiple workers
  # =======================
  redis:
    image: redis:7.0.13
    restart: always
    ports:
      - "6379:6379"
    volumes:
       - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/redis-data:/data

  redis-commander:
    # Visualization tool for redis
    image: rediscommander/redis-commander:latest
    restart: always
    ports:
      - "9092:8081"
    environment:
      - REDIS_HOSTS=local:redis:6379


  # =======================
  # Vector Store Services
  # =======================
  etcd:
    container_name: milvus-etcd
    image: quay.io/coreos/etcd:v3.5.17
    environment:
      - ETCD_AUTO_COMPACTION_MODE=revision
      - ETCD_AUTO_COMPACTION_RETENTION=1000
      - ETCD_QUOTA_BACKEND_BYTES=4294967296
      - ETCD_SNAPSHOT_COUNT=50000
    volumes:
      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/etcd:/etcd
    command: etcd -advertise-client-urls=http://127.0.0.1:2379 -listen-client-urls http://0.0.0.0:2379 --data-dir /etcd
    healthcheck:
      test: ["CMD", "etcdctl", "endpoint", "health"]
      interval: 30s
      timeout: 20s
      retries: 3

  minio:
    container_name: milvus-minio
    image: minio/minio:RELEASE.2024-11-07T00-52-20Z
    environment:
      MINIO_ACCESS_KEY: minioadmin
      MINIO_SECRET_KEY: minioadmin
    ports:
      - "9011:9011"
      - "9010:9010"
    volumes:
      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/minio:/minio_data
    command: minio server /minio_data --console-address ":9011" --address ":9010"
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:9010/minio/health/live"]
      interval: 30s
      timeout: 20s
      retries: 3

  milvus:
    container_name: milvus-standalone
    image: milvusdb/milvus:v2.4.15-gpu
    command: ["milvus", "run", "standalone"]
    environment:
      ETCD_ENDPOINTS: etcd:2379
      MINIO_ADDRESS: minio:9010
      KNOWHERE_GPU_MEM_POOL_SIZE: 2048;4096
    volumes:
      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/milvus:/var/lib/milvus
    ports:
      - "19530:19530"
      - "9091:9091"
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:9091/healthz"]
      interval: 30s
      start_period: 90s
      timeout: 20s
      retries: 3
    depends_on:
      etcd:
        condition: service_healthy
      minio:
        condition: service_healthy
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              capabilities: ["gpu"]
              device_ids: ['${VECTORSTORE_GPU_DEVICE_ID:-0}']


networks:
  default:
    name: nvidia-rag