JMeter API Performance Tests

Small hands-on performance testing project using Apache JMeter against a local FastAPI application.

The project demonstrates HTTP request testing, CSV parameterization, assertions, configurable load generation, non-GUI execution, JTL/HTML reporting, and basic HTTP(S) request recording.

Project Structure

jmeter-api-performance-tests/
├── app/
│   ├── __init__.py
│   └── main.py
├── test-data/
│   └── items.csv
├── test-plans/
│   ├── api-smoke-test.jmx
│   ├── api-performance-test.jmx
│   └── api-recorder-demo.jmx
├── .gitignore
├── requirements.txt
└── README.md

Test Application

The included FastAPI application provides:

• GET /health - basic health check
• GET /items/{item_id} - returns test item data
• GET /slow?delay_ms=100 - introduces a controlled asynchronous delay for performance testing

The /slow endpoint uses asyncio.sleep() to simulate predictable I/O-style waiting without deliberately consuming CPU.

JMeter Test Plans

API Smoke Test

test-plans/api-smoke-test.jmx

Demonstrates:

• CSV Data Set Config
• request parameterization
• HTTP response-code assertion
• JSON JMESPath response assertion
• reusable HTTP Request Defaults

Test data is loaded from:

test-data/items.csv

The test validates three item responses against their expected names.

API Performance Test

test-plans/api-performance-test.jmx

Runs concurrent requests against the controlled /slow endpoint.

The workload can be configured through JMeter properties:

threads
rampup
loops

Default values:

threads = 10
rampup = 1
loops = 20

Example CLI run:

jmeter -n \
  -t test-plans/api-performance-test.jmx \
  -Jthreads=10 \
  -Jrampup=1 \
  -Jloops=20 \
  -l results.jtl

On Windows, use jmeter.bat instead of jmeter if JMeter is not available directly through PATH.

An HTML dashboard can also be generated:

jmeter -n \
  -t test-plans/api-performance-test.jmx \
  -Jthreads=10 \
  -Jrampup=1 \
  -Jloops=20 \
  -l results.jtl \
  -e \
  -o report

This creates:

results.jtl        # raw test results
report/index.html  # main HTML dashboard

The paths are relative to the directory where the command is executed. The report output directory should be empty or not yet exist before generation.

HTTP(S) Recorder Demo

test-plans/api-recorder-demo.jmx

Demonstrates basic use of JMeter's HTTP(S) Test Script Recorder.

A request to the local /slow endpoint was routed through JMeter's proxy on port 8888, producing a reusable HTTP Request sampler and captured HTTP headers.

The recorder is intended for creating test-script requests from observed HTTP traffic; performance measurements are collected when the recorded sampler is subsequently executed.

Example Performance Comparison

A simple local comparison was performed against:

GET /slow?delay_ms=100

Workload	Samples	Average	Median	Throughput
1 concurrent user	20	~109 ms	~105 ms	~9 req/s
10 concurrent users	200	~104 ms	~103 ms	~70 req/s

The response time remained close to the controlled 100 ms delay while throughput increased substantially as requests were handled concurrently.

These results are illustrative rather than a production benchmark. JMeter and the FastAPI application were running on the same physical machine, so operating-system, networking, JVM, Python, and load-generator resource usage can all affect the measurements.

Raw .jtl files and generated HTML reports are intentionally excluded from version control.

Setup

Create and activate a Python virtual environment, then install the application dependencies:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Start the local API:

uvicorn app.main:app --host 127.0.0.1 --port 8001

Verify it:

curl http://127.0.0.1:8001/health

Apache JMeter 5.6.3 was used while developing the test plans.

Performance Testing Concepts Practiced

• concurrent virtual users
• thread ramp-up
• request loops
• response time and latency
• throughput
• percentile-based analysis
• warm-up and connection reuse effects
• load-generator resource limitations
• non-GUI test execution
• HTTP request recording