本地Docker部署Flask/Gunicorn ML应用请求连接被拒绝求助

问题背景 #

我在课程中尝试用Flask/Gunicorn结合Docker本地部署机器学习模型。执行docker run -it --platform linux/amd64 --rm homework_test启动容器后，Gunicorn日志显示正常监听http://0.0.0.0:9696，但在另一个终端运行test.py发送POST请求到http://localhost:9696/predict时，出现连接被拒绝错误。已确认端口匹配，附上相关代码和日志请求排查。

容器启动日志 #

[2024-06-16 10:04:10 +0000] [1] [INFO] Starting gunicorn 22.0.0
[2024-06-16 10:04:10 +0000] [1] [INFO] Listening at: http://0.0.0.0:9696 (1)
[2024-06-16 10:04:10 +0000] [1] [INFO] Using worker: sync
[2024-06-16 10:04:10 +0000] [8] [INFO] Booting worker with pid: 8

test.py执行错误日志 #

(base) marcusleiwe@Marcuss-iMac homework % python test.py
Traceback (most recent call last):
  File "/opt/anaconda3/lib/python3.11/site-packages/urllib3/connection.py", line 203, in _new_conn
    sock = connection.create_connection(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/anaconda3/lib/python3.11/site-packages/urllib3/util/connection.py", line 85, in create_connection
    raise err
  File "/opt/anaconda3/lib/python3.11/site-packages/urllib3/util/connection.py", line 73, in create_connection
    sock.connect(sa)
ConnectionRefusedError: [Errno 61] Connection refused

The above exception was the direct cause of the following exception:

...（省略中间错误栈）
requests.exceptions.ConnectionError: HTTPConnectionPool(host='localhost', port=9696): Max retries exceeded with url: /predict (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x10fb438d0>: Failed to establish a new connection: [Errno 61] Connection refused'))

相关代码 #

Dockerfile

FROM agrigorev/zoomcamp-model:mlops-2024-3.10.13-slim

RUN pip install -U pip #just make sure the pip version is correct
RUN pip install pipenv

WORKDIR /app
COPY ["Pipfile", "Pipfile.lock", "./"]
RUN pipenv install --system --deploy

COPY ["predict.py", "./"]

EXPOSE 9696

ENTRYPOINT ["gunicorn", "--bind=0.0.0.0:9696", "predict:app"]

predict.py（节选）

import pickle 
from flask import Flask, request, jsonify 
import pandas as pd 
import os 
with open('model.bin', 'rb') as f_in: 
     dv, model = pickle.load(f_in) 

app = Flask('duration-prediction') 
@app.route('/predict', methods=['POST']) 
def predict(): 
    .... 
    y_pred = model.predict(X_val) 
    df_result = pd.DataFrame({ 'predicted_duration': y_pred }) 
    results = { 'mean_duration': df_result['predicted_duration'].mean() } 
    return jsonify(results) 

if __name__ == "__main__": 
    app.run(debug=True, host='0.0.0.0', port=9696)

test.py

import requests

year_month = {
    "YEAR" : 2023,
    "MONT": 5
}

url = 'http://localhost:9696/predict' 
requests.post(url, json=year_month)

解决步骤 #

1. 添加Docker端口映射
   你的启动命令没有将容器内部的9696端口映射到宿主机，导致宿主机无法访问容器内的服务。修改启动命令为：

   docker run -it --platform linux/amd64 -p 9696:9696 --rm homework_test
   

   -p 9696:9696参数实现了宿主机端口与容器端口的映射，此时宿主机的localhost:9696就能指向容器内的服务。

2. 验证容器内服务可用性（可选）
   如果添加映射后仍有问题，可进入容器内部测试服务是否正常：

   # 获取容器ID
   docker ps
   # 进入容器并发送请求
   docker exec -it <容器ID> curl -X POST http://localhost:9696/predict -d '{"YEAR":2023,"MONT":5}' -H "Content-Type: application/json"
   

   若容器内可正常返回结果，说明问题出在宿主机与容器的端口映射；若容器内也无法访问，再检查Gunicorn启动参数或Flask应用逻辑。

3. 检查宿主机端口占用
   确认宿主机的9696端口是否被其他进程占用：

   # macOS/Linux
   lsof -i :9696
   # Windows
   netstat -ano | findstr :9696
   

   如果端口被占用，可停止占用进程，或修改Docker映射的宿主机端口（如-p 9697:9696），同时更新test.py中的请求URL为http://localhost:9697/predict。

内容的提问来源于stack exchange，提问作者Marcus Leiwe