pytest-fastprom

Proof of concept

This project is a proof of concept. The approach still needs real-world feedback (REX) before it can be considered validated, so treat results as a signal, not a guarantee. Most of the code was generated using generative AI.

Why this project

pytest-fastprom started as a proof of concept around a specific blind spot: performance regressions you can't see in a code review.

The kind of change that looks harmless in a diff but quietly costs you milliseconds at runtime:

• an ORM relationship that now triggers an N+1 or a cache-busting join,
• a new middleware that runs on every request,
• an algorithm swapped for one with worse complexity on real-world input.

None of these break a test. None of them show up until latency creeps into production. The idea here is to catch those drawbacks early, right in the test suite, by reusing the measurement tool your app already trusts: Prometheus.

Not a load-testing tool

This is not a replacement for real load testing. Reach for k6 or Locust when you need concurrency, sustained throughput, or production-scale traffic.

`pytest-fastprom` aims for something narrower and faster: a quick signal, on every test run, that a change didn't silently regress the latency or error budget you promised, measured with the same Prometheus histograms you already ship.

Installation

uv add --dev pytest-fastprom
# or
pip install pytest-fastprom