需求:对接口api/mocktest 接口进行流式返回的mock,其中要求接口请求方式为:api/mocktest/{first_time}/{total_time},first_time为首token耗时、total_time为流式返回总耗时
import sys
from flask import Flask, Response, request
import json
import time
app = Flask(__name__)
@app.route("/api/mocktest/<int:first_time>/<int:total_time>", methods=["POST"])
def process(first_time, total_time):
if first_time >= total_time:
return Response(json.dumps({"error": "First time is greater than total time"}), status=400)
"""
/api接口定义,动态获取first token time和complete response time
"""
stream = request.json.get("stream")
stream_num = 10
# 根据请求参数计算平均时间
avg_time = (total_time - first_time) / (stream_num - 1)
# 初始化流式数据
mock_stream_data = [[json.dumps({"data": "this is a stream mock1", "status": 0}), first_time]]
for i in range(2, stream_num + 1):
res = [json.dumps({"data": "this is a stream mock" + str(i), "status": 0}), avg_time]
mock_stream_data.append(res)
# 初始化非流式数据
no_stream_time = first_time
mock_nostream_data = [
[json.dumps({"data": "this is a nostream mock", "status": 0}), no_stream_time]
]
def generate_stream():
"""
流式返回数据
"""
for each in mock_stream_data:
v = each[0]
t = each[1]
if t > 0:
time.sleep(t) # 等待指定的时间后再发送下一条数据
yield v + "\r\n"
sys.stdout.flush()
def generate_nostream():
"""
非流式返回数据
"""
v = mock_nostream_data[0][0]
t = mock_nostream_data[0][1]
if t > 0:
time.sleep(t)
yield v + "\r\n"
if stream:
return Response(generate_stream(), mimetype="text/plain")
else:
return Response(generate_nostream(), mimetype="text/plain")
app.run()
python运行后通过curl命令访问地址:
流式请求:
curl --location '127.0.0.1:5000/api/mocktest/3/10' \
--header 'Content-Type: application/json' \
--data '{
"messages": [{"role": "user", "content": "发起请求"}],
"stream": true
}'
结果:
非流式请求:
curl --location '127.0.0.1:5000/api/mocktest/3/10' \
--header 'Content-Type: application/json' \
--data '{
"messages": [{"role": "user", "content": "发起请求"}],
"stream": false
}'
结果: