Databricks Apps & 종합 가이드

command:
  - "streamlit"
  - "run"
  - "app.py"
  - "--server.port"
  - "8000"
  - "--server.address"
  - "0.0.0.0"

env:
  - name: "DATABRICKS_WAREHOUSE_ID"
    valueFrom: "warehouse-id"
  - name: "SERVING_ENDPOINT"
    value: "mlops-agent-endpoint"

resources:
  - name: "warehouse-id"
    type: "sql_warehouse"
    sql_warehouse:
      id: "abc123def456"
      permission: "CAN_USE"
  - name: "serving-endpoint"
    type: "serving_endpoint"
    serving_endpoint:
      name: "mlops-agent-endpoint"
      permission: "CAN_QUERY"

command:
  - "uvicorn"
  - "main:app"
  - "--host"
  - "0.0.0.0"
  - "--port"
  - "8000"

env:
  - name: "DATABRICKS_WAREHOUSE_ID"
    valueFrom: "warehouse-id"

resources:
  - name: "warehouse-id"
    type: "sql_warehouse"
    sql_warehouse:
      id: "abc123def456"
      permission: "CAN_USE"

# 1. 앱 생성 (최초 1회)
databricks apps create \
  --name mlops-agent-app \
  --description "MLOps 모니터링 및 Agent 채팅"

# 2. 앱 배포 (코드 변경 시마다)
databricks apps deploy mlops-agent-app \
  --source-code-path ./app

# 3. 배포 상태 확인
databricks apps get mlops-agent-app

# 4. 로그 확인
databricks apps get-logs mlops-agent-app

# Before (로컬 개발)
from databricks.sdk import WorkspaceClient
w = WorkspaceClient(
    host="https://xxx.cloud.databricks.com",
    token="dapi..."  # PAT 하드코딩
)

# After (Databricks Apps)
from databricks.sdk import WorkspaceClient
w = WorkspaceClient()  # 서비스 프린시펄 자동 인증

import os

# 하드코딩 제거, 환경 변수 사용
warehouse_id = os.environ.get("DATABRICKS_WAREHOUSE_ID")
endpoint_name = os.environ.get("SERVING_ENDPOINT", "default-agent")

streamlit>=1.38.0
databricks-sdk>=0.30.0
plotly>=5.0.0

Agent UI 기술 선택
│
├─ "2주 안에 데모를 보여줘야 한다"
│   └─→ Streamlit (채팅+대시보드) 또는 Chainlit (Agent 추론 시각화)
│
├─ "Hugging Face 모델을 데모하고 싶다"
│   └─→ Gradio
│
├─ "차트 중심 모니터링 대시보드를 만든다"
│   └─→ Dash
│
├─ "기존 웹 포털에 Agent를 통합해야 한다"
│   └─→ FastAPI (백엔드) + 기존 프론트엔드
│
├─ "Databricks Apps에 배포해야 한다"
│   ├─ 채팅 UI 중심 → Streamlit
│   ├─ 대시보드 중심 → Dash
│   └─ API 서버 → FastAPI
│
└─ "사용자 100명+, 커스텀 UX 필수"
    └─→ FastAPI + React/Next.js

# 이 코드만으로 Agent 채팅 앱 완성 (약 20줄)
import streamlit as st
from databricks.sdk import WorkspaceClient

st.title("My Agent")
w = WorkspaceClient()

if "messages" not in st.session_state:
    st.session_state.messages = []

for msg in st.session_state.messages:
    with st.chat_message(msg["role"]):
        st.markdown(msg["content"])

if prompt := st.chat_input("질문하세요"):
    st.session_state.messages.append({"role": "user", "content": prompt})
    with st.chat_message("user"):
        st.markdown(prompt)

    response = w.serving_endpoints.query(
        name="my-agent-endpoint",
        messages=st.session_state.messages
    )
    result = response.choices[0].message.content

    with st.chat_message("assistant"):
        st.markdown(result)
    st.session_state.messages.append({"role": "assistant", "content": result})

import streamlit as st
from databricks.sdk import WorkspaceClient

w = WorkspaceClient(
    host="https://my-workspace.cloud.databricks.com",
    token="dapi1234567890abcdef"
)

response = w.serving_endpoints.query(
    name="my-agent",
    messages=[{"role": "user", "content": "안녕"}]
)
st.write(response.choices[0].message.content)