0

Hi I am using AWS EC2 instance (c5.18xlarge (72vcpu and 144 Gib ram)) with Nginx + Django and postgresql. The problem is it gets to 100% CPU utilisation in 400-500 concurrent users and slows down. Here are my settings:

postgresql.conf

max_connections = 800
shared_buffers = 50000MB

Nginx.conf

user www-data;
worker_processes 128;
pid /run/nginx.pid;
include /etc/nginx/modules-enabled/*.conf;
worker_rlimit_nofile 50000;
events {
        worker_connections 10000;
        #multi_accept on; 
}

sysctl.conf

net.core.somaxconn = 65000
net.core.netdev_max_backlog = 65535
fs.file-max = 90000
net.ipv4.ip_forward = 1
kernel.shmmax=1006632960

gunicorn_start.bash

NAME="django_app" # Name of the application

DJANGODIR=/home/ubuntu/project              # Django project directory
SOCKFILE=/home/ubuntu/env/run/gunicorn.sock  # we will communicte using this unix socket
USER=ubuntu                                         # the user to run as
GROUP=ubuntu                                        # the group to run as
NUM_WORKERS=128                                      # how many worker processes should Gunicorn spaw
TIMEOUT=120
DJANGO_SETTINGS_MODULE=project.settings      # which settings file should Django use
DJANGO_WSGI_MODULE=project.wsgi              # WSGI module name
echo "Starting $NAME as `whoami`"

I know the server can't overload on the current usage of the app. Please suggest how can this be solved. Or how can it be debugged to reach the problem.

I am also attaching the top screenshot: enter image description here

Prashant Pandey
  • 221
  • 3
  • 16
  • You have to figure out what uses the CPU time and why. That is called "profiling". – Laurenz Albe Mar 09 '21 at 07:40
  • using "top" what process is using all the CPU? – Trent Mar 11 '21 at 00:41
  • @Trent I have attached the screen shot. – Prashant Pandey Mar 12 '21 at 07:56
  • Without knowing what the endpoints are doing (CPU activities) - will be hard to debug remotely. Given the power of that server, there is no way 400-500 concurrent users should make a dent on resources, unless you have a bug in the code somewhere which causes an infinite loop, or does something daft causing the CPU to get smashed. – Trent Mar 12 '21 at 11:35

0 Answers0