Session 1 · Why load testing, and installing Locust#

The big idea#

A website that works for you alone may fall over when a thousand people use it at once. Load testing means sending realistic traffic at your service on purpose, before real users do, and measuring what happens: how fast it answers, how many requests per second it handles, and when it starts to fail. Locust is a free tool where you describe the users in plain Python.

Everyday example: a wedding hall booked for 500 guests. You would not wait for the wedding day to find out that the kitchen can only serve 200 plates an hour. A caterer does a trial run with 500 imaginary guests first, notes where the queue forms, and fixes it. Load testing is that trial run for your server.

Locust runs many simulated users against your service while you watch response time, requests per second and failures.

The questions a load test answers#

Can the service handle the expected traffic (say, 500 users at once)?
How does performance degrade as traffic grows?
Where is the breaking point, and what breaks first (the app, the database, the network)?

Real examples: a shop before a sale day, a streaming service before a launch, a trading app at market open. In each case a slow checkout or a dropped video is lost money.

The numbers to watch#

Metric	Meaning	Rule of thumb
Response time	How long one request takes	Many sites aim for under 2 s
Throughput (requests/s)	How many requests are served per second	Higher is better, until errors start
Error rate	Share of requests that fail	Should stay near 0%
Percentiles (p95, p99)	The time that 95% or 99% of requests beat	Shows the slow tail that averages hide
Concurrent users	How many users act at the same time	The load you chose

An average of 200 ms can hide the fact that 1 request in 20 takes 5 seconds. Always look at p95 and p99 too.

Install Locust#

pip install locust
locust --version

locust 2.28.0 from <site-packages>/locust

Use a virtual environment (python -m venv .venv) if you like. The version above is what every example in this course was run with. Locust has two ways to run: with a web page (locust -f file.py, then open http://localhost:8089) or headless from the command line, which this course uses because it prints results you can read.

The practice shop#

All sessions test one small web shop that runs on your own computer, so you can load test freely without harming anyone. code/shop_server.py uses only Python’s standard library:

python shop_server.py        # listens on http://127.0.0.1:8765

URL	Does
`GET /products`, `GET /products/<id>`	Product list and detail
`POST /login`	Returns a token for user `demo` / password `secret`
`GET /profile`	Needs the token
`POST /orders`	Places an order
`GET /slow`	Takes 0.3 s on purpose
`GET /flaky`	Fails with a 500 every fourth time

Common mistakes#

Load testing someone else’s website. That can look like an attack and may be illegal. Test your own systems, or ones you have written permission for.
Testing the laptop instead of the service. If your computer is the bottleneck, the numbers mean nothing. Watch your own CPU (Session 8 shows how to add workers).
Reading only the average. Check percentiles and the error rate.
Testing on production by surprise. Use a staging copy, or warn the team and pick a quiet time.
One test, then done. Performance changes with every release; repeat the test regularly.

Hands-on exercises#

Exercise 1. Install Locust in a virtual environment and print its version.

Solution

python -m venv .venv
source .venv/bin/activate
pip install locust
locust --version

You see a line like locust 2.28.0. Your version may be newer.

Exercise 2. Start the practice shop and call one URL with curl.

Solution

python shop_server.py &
curl -s http://127.0.0.1:8765/products/2

It prints {"id": 2, "name": "Phone", "price": 500}. Stop the server later with kill %1.

Exercise 3. A service averages 200 ms but its p99 is 6 seconds. Is it healthy? Why or why not?

Solution

Not necessarily. 1 request in 100 takes about 6 seconds, which a busy user will meet often (a page may make dozens of requests). The average hides this tail; the p95 and p99 reveal it.

Practice#

Take the Session 1 quiz. The original Why Load Testing? and Locust Installation pages are in the reference section.