Load testing Flask API

This repository contains reports of load capacity of a simple Flask API deployed using different configurations:

1. Single instance: py app.py The most basic form of flask server running in your local machine.

2. Using a load balancer built with flask: It distributes the load in a random manner.

3. Using a load balancer built with flask: It distributes the load in a cyclic manner (Similar to round robin algorithm).

4. Gunicorn: Using 5 gunicorn workers. The number of workers is decided by 2 $\times$ Cores + 1.

5. Gunicorn: Using 10 gunicorn workers. It is twice the recomeneded value.

How to run

First clone the repository:

git clone https://github.com/Sriram-bb63/maximising-flask-performance.git

To run a single instance:

py instances/app1.py

To run the load balancer:

chmod +x run.sh
./run.sh

py load_balancer.py

To change between cyclic mode and random mode, open load_balancer.py and comment/uncomment the respective code blocks.

To run using Gunicorn:

gunicorn instances.app1:app

Traffic

• Max concurrent users: 1000
• User spawn rate: 100
• Time period: 1 minute

Result

	Total requests	Requests/sec	Failures/sec
Single instance	9871	164.3	1.4
Load balancer (Random URL)	4788	79.7	4.5
Load balancer (Cyclic URL)	4125	68.8	3.7
Gunicorn (5 workers [Recommended for 2 cores])	16335	271.8	0.0
Gunicorn (10 workers [2x recommened value])	13443	223.8	0.0

Detailed reports

• Single instance
• Reverse proxy (Random)
• Reverse proxy (Cyclic)
• Gunicorn (5 workers [Recommended for 2 cores])
• Gunicorn (10 workers [2x recommended workers])