Skip to content

README.md

Latest commit

 

History

History
45 lines (34 loc) · 1.6 KB

README.md

File metadata and controls

45 lines (34 loc) · 1.6 KB
Raw

LearnLoop load tests

The Locust scenario exercises the deterministic study-history path so the local result measures LearnLoop and SQLite rather than OpenAI latency:

  • read-heavy: history, search, and all dashboard analytics endpoints
  • write-heavy: POST /quiz_results, followed by GET /quiz_results/:id

Generation endpoints are intentionally excluded. They require an external OpenAI key and would measure provider availability and model latency as well as the application. They need a separate, explicitly configured provider test.

Run locally

From the repository root, install the pinned load-test dependency:

python3.11 -m pip install -r load_tests/requirements.txt

Start the backend in another terminal with a load-test database and the production-style local server:

SUPABASE_DB_URI=sqlite:////tmp/learnloop-load-test.db \
  gunicorn --chdir backend --bind 127.0.0.1:5050 \
  --workers 4 --threads 2 --timeout 30 wsgi:app

Run 500 users with a short ramp and a bounded duration:

locust -f load_tests/locustfile.py --headless \
  --host http://127.0.0.1:5050 \
  --users 500 --spawn-rate 50 --run-time 2m \
  --csv docs/load-tests/locust-500-users-gunicorn \
  --html docs/load-tests/locust-500-users-gunicorn.html \
  --only-summary

The CSV and HTML files are the raw report artifacts. Save each run under a unique prefix. The Markdown report must record total requests, failures and failure rate, requests per second, p50, p95, the exact command, server configuration, and observed SQLite lock errors. Do not replace a failed run with a successful rerun in the same report.