Flask超高并发,Flask多进程gevent ,Flask非gunicorn gevent多进程
1. flask + gevent + multiprocess + wsgi简介
常常大家都是用gunicorn来解决flask后端部署并发的问题, 然而觉得自启多进程是为更优雅的高并发方式。这样就不需要gunicorn了。也没有额外的第三方部署工作,于是有了以下flask + gevent + multiprocess + wsgi的测试
2. flask + gevent + multiprocess + wsgi程序代码app.py
# coding: utf-8
# flask + gevent + multiprocess + wsgi
from gevent import monkey
from gevent.pywsgi import WSGIServer
monkey.patch_all()
import datetime
import os
from multiprocessing import cpu_count, Process
from flask import Flask, jsonify
app = Flask(__name__)
@app.route("/cppla", methods=['GET'])
def function_benchmark():
return jsonify(
{
"status": "ok",
"time": datetime.datetime.now().strftime('%Y-%m-%d %H:%M'),
"pid": os.getpid()
}
), 200
def run(MULTI_PROCESS):
if MULTI_PROCESS == False:
WSGIServer(('0.0.0.0', 8080), app).serve_forever()
else:
mulserver = WSGIServer(('0.0.0.0', 8080), app)
mulserver.start()
def server_forever():
mulserver.start_accepting()
mulserver._stop_event.wait()
for i in range(cpu_count()):
p = Process(target=server_forever)
p.start()
if __name__ == "__main__":
# 单进程 + 协程
run(False)
# 多进程 + 协程
# run(True)
3. 测试环境
机器配置
4vCPU
8GB
环境配置
Centos 8
Python 3.6.8
Flask 1.1.2
Gevent 1.5.0
压测工具
WRK
4H8G机器
4. 单进程 + 协程 + WSGI
4.1 测试结果
QPS : 2361 r/s
4.2 测试详情
[root@vm5 ~]# nohup python3 cppla.py &
[1] 9371
[root@vm5 ~]# nohup: ignoring input and appending output to 'nohup.out'
[root@vm5 ~]# ps -ef | grep cppla
root 9371 4184 4 03:33 pts/0 00:00:00 python3 cppla.py
root 9377 4184 0 03:33 pts/0 00:00:00 grep --color=auto cppla
[root@vm5 ~]#
[root@vm4 wrk]# curl 10.10.10.5:8080/cppla
{"pid":9371,"status":"ok","time":"2020-04-20 04:19"}
[root@vm4 wrk]# curl 10.10.10.5:8080/cppla
{"pid":9371,"status":"ok","time":"2020-04-20 04:19"}
[root@vm4 wrk]# curl 10.10.10.5:8080/cppla
{"pid":9371,"status":"ok","time":"2020-04-20 04:19"}
[root@vm4 wrk]# wrk -t12 -c400 -d30s http://10.10.10.5:8080/cppla
Running 30s test @ http://10.10.10.5:8080/cppla
12 threads and 400 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 164.40ms 21.15ms 515.30ms 80.85%
Req/Sec 199.61 47.81 565.00 70.38%
71237 requests in 30.10s, 10.95MB read
Requests/sec: 2366.72
Transfer/sec: 372.44KB
[root@vm4 wrk]# wrk -t20 -c800 -d30s http://10.10.10.5:8080/cppla
Running 30s test @ http://10.10.10.5:8080/cppla
20 threads and 800 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 329.39ms 93.86ms 1.78s 92.80%
Req/Sec 126.19 80.69 696.00 67.71%
71075 requests in 30.10s, 10.92MB read
Socket errors: connect 0, read 0, write 0, timeout 97
Requests/sec: 2361.39
Transfer/sec: 371.65KB
[root@vm4 wrk]#
5. 多进程 + 协程 + WSGI
5.1 测试结果
QPS : 7500 r/s
5.2 测试详情
[root@vm5 ~]# nohup python3 cppla.py &
[1] 9537
[root@vm5 ~]# nohup: ignoring input and appending output to 'nohup.out'
[root@vm5 ~]# ps -ef | grep cppla
root 9537 4184 5 04:32 pts/0 00:00:00 python3 cppla.py
root 9542 9537 0 04:32 pts/0 00:00:00 python3 cppla.py
root 9543 9537 0 04:32 pts/0 00:00:00 python3 cppla.py
root 9544 9537 0 04:32 pts/0 00:00:00 python3 cppla.py
root 9545 9537 0 04:32 pts/0 00:00:00 python3 cppla.py
root 9547 4184 0 04:32 pts/0 00:00:00 grep --color=auto cppla
[root@vm5 ~]#
[root@vm4 wrk]# curl 10.10.10.5:8080/cppla
{"pid":9543,"status":"ok","time":"2020-04-20 04:34"}
[root@vm4 wrk]# curl 10.10.10.5:8080/cppla
{"pid":9542,"status":"ok","time":"2020-04-20 04:34"}
[root@vm4 wrk]# curl 10.10.10.5:8080/cppla
{"pid":9545,"status":"ok","time":"2020-04-20 04:34"}
[root@vm4 wrk]# wrk -t12 -c400 -d30s http://10.10.10.5:8080/cppla
Running 30s test @ http://10.10.10.5:8080/cppla
12 threads and 400 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 56.10ms 15.16ms 187.30ms 85.05%
Req/Sec 590.77 79.95 830.00 67.97%
212138 requests in 30.08s, 32.60MB read
Requests/sec: 7051.89
Transfer/sec: 1.08MB
[root@vm4 wrk]# wrk -t20 -c800 -d30s http://10.10.10.5:8080/cppla
Running 30s test @ http://10.10.10.5:8080/cppla
20 threads and 800 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 101.59ms 40.23ms 337.80ms 66.06%
Req/Sec 394.20 109.48 0.97k 74.47%
235844 requests in 30.10s, 36.25MB read
Requests/sec: 7835.77
Transfer/sec: 1.20MB
[root@vm4 wrk]#
补充:4核8线程3.6GHZ物理机上多进程并发高达15000QPS
也就是“理想”情况,每秒1w5 QPS,每小时5400万请求。
[root@vm4 wrk]# wrk -t30 -c1000 -d30s http://10.10.10.1:8080/cppla
Running 30s test @ http://10.10.10.1:8080/cppla
30 threads and 1000 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 67.67ms 18.54ms 312.76ms 81.81%
Req/Sec 489.92 76.58 710.00 68.65%
440105 requests in 30.09s, 68.07MB read
Requests/sec: 14626.25
Transfer/sec: 2.26MB
[root@vm4 wrk]#
参考链接:
1. https://www.cplusplus.me/2984.html