App runs slow when multiple users accessing

šŸŽˆ Using Streamlit
,
1

Hi,

Similar to the topic discussed in [App runs slow when multiple users accessing simultaneously], my app is being used by 20 people.

I am facing an issue where the response speed of my program gradually slows down after approximately 5 days of continuous usage. Clicking a button takes about 1-3 seconds. However, upon restarting the program, the response time significantly improves to 10 ms.

Is there any suggestion to me?
anyone, helpā€¦

2

Knowing nothing specific about your app, my first guess would be that you have some kind of memory leak or something. Iā€™m not sure if anyone can provide any meaningful tips without knowing more about your app and its code.

For general advice, try profiling your appā€™s memory usage to see if you can gain insight into what may be contributing to the digital clutter.

1 Like
3

hi,
i think there are nothing about code, i try to use below code:

import streamlit as st 
a=st.radio('sssss',['1','2'])
if a=='1':
    st.write('1')
else:
    st.write('2')

Then i use siege to test, when the concurrence up to 2, pageā€™s response is slower than 1.
if i set the concurrence to 10 or more, the response time will be long, >5 seconds.

I run my streamlit on a windows 2022 server, i think its CPU and Memory is enough.

Never tried on linux.

thanks for reply but i still donā€™t know the problem.

4

I ran your code in a Windows 10 computer and then opened > 15 sessions in the same computer without experiencing any noticeable delays. So, if you see a lack of responsiveness when connecting from other computers, the issue must be network latency. How the number of sessions is related to network latency, I donā€™t know.

I tested manually, not with siege.

5

thanks for your reply agian, i guess this issue will only happend when multiple computers connect to the sever. I think my network it good, i even try it under local network. I only know little about the network, but i try to use nginx to proxy my app, i run multiple app.py on different ports to avoid above problem. But there are some new problemsā€¦Hard to solve.
Here is my code of nginx.conf:

worker_processes  1;

events {
    worker_connections  1024;
}

http {
    upstream streamlit_servers {
        server 192.168.1.5:8880;
        server 192.168.1.5:8881;
        server 192.168.1.5:8882;
		server 192.168.1.5:8883;
		server 192.168.1.5:8884;
		server 192.168.1.5:8885;
		server 192.168.1.5:8886;
		server 192.168.1.5:8887;
    }

    server {
        listen 8888;
        server_name localhost;

        location / {
            proxy_pass http://streamlit_servers;
            proxy_http_version 1.1;
            proxy_set_header Upgrade $http_upgrade;
            proxy_set_header Connection "Upgrade";
            proxy_next_upstream error timeout invalid_header http_500 http_502 http_503 http_504;
        }
    }
}
6

@WMKTT1 could you please let us know what version of streamlit are you running and if you tried to turn-off fast reruns?

7

Hi TomJohn:
I use the newest streamlit, I tried to test a simple script again yesterday using seige and Selenium. I found that when I used ā€œseige -c 4ā€, which means 4 concurrent connections, there was a noticeable delay when clicking on ā€œst.radioā€, but it was not as obvious when using ā€œ-c 2ā€. However, the experience was slightly better when using Selenium. Now I use nginx as a reverse proxy, which has solved the problem. Please forgive me if I have any misunderstandings about streamlit due to my limited knowledge. However, I still want to know, theoretically, how many people can click at the same time if I only use ā€œstreamlit run simple.pyā€, where ā€œ.pyā€ is a very simple script? The loading issue I encountered before was when about 12 people clicked at the same time.

8

I donā€™t think there is a theoretical limit. IIRC you the issue you found was slowness, not a limit. There is a practical limit depending on the system resources available to the application.