在Vertex AI上用FastAPI创建端点时遭遇/health接口404问题求助

1. 修正代码缩进错误 #

Python对缩进敏感，原代码中startup函数内的代码存在缩进混乱，会导致语法错误或路由无法正确注册。修正后的startup函数如下：

@app.on_event("startup")
def startup():
    path = os.environ["AIP_STORAGE_URI"]
    path = os.path.join(path, model_file)
    logger.info(f"Loading model state from: {path}")

    if path.startswith("gs://"):
        destination_file_name = f"/tmp/{model_file}"
        storage_client = storage.Client()
        with open(destination_file_name, "wb") as file:
            # Download file from GCP
            storage_client.download_blob_to_file(path, file)
    else:
        logger.info("Local path...")
        destination_file_name = path
    
    model_files = joblib.load(destination_file_name)
    global_items["model"] = model_files

    logger.info("State successfully retrieved from GCP")

2. 修正主模块入口判断 #

原代码中if __name__ == "__name__"是错误的，会导致uvicorn启动逻辑无法触发，需改为：

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=8080, reload=False, log_level="info")

3. 确认健康检查端点的正确性 #

• 确保/health路由是全局级别的定义，没有被错误缩进包裹在其他函数内部。
• 检查是否导入了必要的依赖模块：

from fastapi import FastAPI, Response, status
import os
import storage
import joblib
import uvicorn
from logging import getLogger

logger = getLogger(__name__)
app = FastAPI()
global_items = {}
model_file = "your_model_filename.joblib"  # 替换为实际模型文件名

4. 验证Vertex AI部署配置 #

• Vertex AI默认使用GET /health进行健康检查，无需额外配置，确保服务监听端口为8080（代码中已设置）。
• 本地测试：运行代码后执行curl http://localhost:8080/health，确认返回200状态码后再重新构建镜像部署。

内容的提问来源于stack exchange，提问作者Moustafa Ashraf