load-testing

This is a website load testing service with the following features

• url and number of requests parameterized
• supports concurrency requests
• calculates 50th, 75th, 95th, 99th percentile latencies
• requests made through a POST endpoint

After running a load test, the endpoint returns

• total requests completed
• concurrency
• total time for requests
• average latency
• percentiles

Architecture diagram

This program is hosted on GCP's cloud run with autoscaling. CI/CD pipeline is managed via cloud build such that whenever code is pushed to main, cloud build packages and containerizes it, and builds and deploys it to cloud run.

Testing locally

1. Clone the repository
2. Setup virtual environment
   • Install virtualenv: pip install virtualenv
   • Create a virtual environment: virtualenv venv
   • Activate virtualenv: Linux/macOS: source venv/bin/activate
3. Install dependencies: pip install -r requirements.txt
4. Adjust sample parameters within top-level code to your request
5. Run the program: python HTTPLoadTest.py

Using the API

The service is hosted on https://load-testing-djjegt4o7q-ue.a.run.app

API Endpoints

HTTP Verbs	Endpoints	Action
POST	/run_load_test	To run load test on a website

POST https://load-testing-djjegt4o7q-ue.a.run.app/run_load_test

Sample Request Body

{
    "url": "https://www.google.com",
    "num_requests": 100,
    "concurrency": 4,
    "request_type": "GET"
}

Sample Response

{
    "average_latency": 0.5338004302978515,
    "concurrency": 4,
    "errors": 0,
    "percentiles": {
        "percentile_50th": 0.5479679107666016,
        "percentile_75th": 0.5910550355911255,
        "percentile_95th": 0.6173677563667297,
        "percentile_99th": 0.795107543468476
    },
    "total_requests": 100,
    "total_time": 13.496866941452026
}

Future improvements

• error handling around invalid requests with accurate HTTP response codes. For this project any errors are returned back as 500 and cloud run re-instantiates the server.
• authenticated endpoint access.
• endpoint DDOS protection by Cloud Armor firewall rules.
• requests saved to DB and then visualized
• Deployment on compute engine VM with multi-region instances connected with a global load balancer. With cloud run, instances can only be deployed in a single region.