Flask cannot even serve 50 requests per second?-CodePudding

I am python user and tried to make python REST API to run a service.
Yet, the stress test showed that the python cannot handle even just 50 users per second!
I guess the error comes out when the nginx cannot assign requested address while the gunicorn workers restarts, but the max time states that such simple function takes 674ms.
I think such high request time is caused by the bottleneck from request queue, but wonder my setting is that much wrong for such simple programme. \

The dummy function code is

@application.route('/dummy', methods = ['POST'])
def helloworldmaker():
    return jsonify('Hello World')

With the function I ran Jmeter test and the settings, results and environments are:

SETTING
Number of Threads: 50
Ramp-up priod: 1
Loop count: 1000

Result
Average(ms): 13
Min(ms): 2
Max(ms): 674
Error(%): 4.36% (99: Cannot assign requested address)
Throughput: 2531.8/sec

my server specification
server name: HPE ProLiant DL360 Gen10
Number of CPU: 40
Number of Cores per CPU: 10
Total Memory/Free Memory: 62G/22G
Total Swap Memory/Free Swap Memory: 15/7.8G

I think the server is not that bad to yield such performance.

Briefly explain the structure, I made Hello World printing API using flask which are passed to wsgi.py.

The wsgi.py passed to my gunicorn service file, where the code is

[Unit]
Description=gunicorn daemon
After=network.target

[Service]
User=root
Group=root
WorkingDirectory=/MycodeFolder
ExecStart=/usr/local/bin/gunicorn --config /etc/systemd/system/myGunicorn.config.py --bind '0.0.0.0:7211' -m 007 wsgi:application


[Install]
WantedBy=multi-user.target

where the gunicorn configuration is

import multiprocessing

workers = multiprocessing.cpu_count() * 2   1
worker_class = 'gevent'
timeout = 300
max_requests = 1000
max_requests_jitter = 50

errorlog = '/var/log/gunicorn_error.log'
accesslog = '/var/log/gunicorn_access.log'

My gunicorn is inside Docker container, and is connected to outer(host) Nginx via proxy pass.

My outer Nginx settings are
SERVER Block

server {
    listen       18445;

    #for dummy test
    location /dummy {
         proxy_set_header Host $host;
         proxy_set_header X-Real-IP $remote_addr;
         proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
         proxy_pass http://dockers/dummy;
    }
}

HTTP Block

user nginx;
worker_processes 10;
error_log /var/log/nginx/error.log;
pid /run/nginx.pid;

include /usr/share/nginx/modules/*.conf;

events {
    worker_connections 3096;
}

http {
    log_format  main  '$remote_addr - $remote_user [$time_local] "$request" '
                      '$status $body_bytes_sent "$http_referer" '
                      '"$http_user_agent" "$http_x_forwarded_for" $request_time';

    access_log  /var/log/nginx/access.log  main;

    sendfile            on;
    tcp_nopush          on;
    tcp_nodelay         on;
    keepalive_timeout   65;
    types_hash_max_size 2048;

    include             /etc/nginx/mime.types;
    default_type        application/octet-stream;

    upstream dockers {
        server 127.0.0.1:7211;
    }
    include /etc/nginx/conf.d/*.conf;
}

Would like to know whether my settings are wrong, or just the python or middleware is not good for real time webserving purpose....
I reckon, because there are some companies which operate services using python, there must be an answer or I made fundamental mistake...
Hope to know the right setting, or check if the whole system is not suitable for webserving purpose!

CodePudding user response：