Turn your SLOs into pytest assertions.

Smoketest your FastAPI against the latency and error budgets you actually promised. Built on the Prometheus metrics your app already exposes, not wall-clock timeit guesses.

def test_home_meets_slo(instrumented_client, metrics):
    instrumented_client.get("/")
    metrics.assert_p99_below(1.0, handler="/")
    metrics.assert_no_server_errors()

Or declare the SLO right on the test:

@pytest.mark.metrics(p99_below=1.0, min_requests=3, no_errors=True)
def test_route_within_budget(instrumented_client):
    for _ in range(3):
        instrumented_client.get("/")

Features

• Prometheus-native. Asserts on real histogram metrics, not stopwatch timing.
• Isolated registry per test. No metric leaks between tests.
• SLO assertions on P50, P99, request counts, and error rates.
• Custom metrics. assert_metric checks any Counter/Gauge/Histogram your app exposes.
• Error budgets. max_error_rate / max_4xx_rate assert a ratio, not just a zero count.
• Regression detection. Save a baseline, fail the build when it drifts.
• Environment-aware baselines. Saved runs record git commit + machine info and warn on cross-machine comparisons.
• Warmup support. Cold starts stay out of your measurement window.

Install

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

Docs

Full guide, marker reference, and regression detection: https://lujeni.github.io/pytest-fastprom/.