Motivation

When you’re processing load test timings (or any latency distribution), you typically want buckets like p50/p95/p99 quickly. percently provides a small, fast Rust implementation exposed to Python.

Features

• Fast: Rust core
• Simple API: one call to compute a percentile
• Python-friendly: works with regular lists/iterables of floats

Installation

Using uv:

uv add percently

Using pip:

pip install percently

🧮 Usage

import percently

data = [10.5, 22.0, 18.7, 19.2, 30.1, 25.3]

# Calculate the 95th percentile (f64)
p95 = percently.percentile(data, 95)
print(f"95th percentile: {p95:.2f}")

Streaming (iterative) percentile estimates

If you have a generator/iterable and want running percentile estimates without loading all values into memory, use stream.

import percently

def timings():
    for i in range(1, 1_000_001):
        yield float(i % 10_000) / 10_000.0

# Yield an updated p95 estimate every 1000 samples (approximate, bounded memory)
for est in percently.stream(timings(), 95, every=1000):
    last = est

print("final approx p95:", last)

Notes:

• stream(..., 0, ...) yields the running min (exact)
• stream(..., 100, ...) yields the running max (exact)
• All values must be finite floats (no NaN/inf)

Development

Build/install locally (editable) with maturin:

uv sync
uv run maturin develop

Releasing

This repo is configured for signed release commits/tags and tag-based publishing:

• Create a signed tag (e.g. 0.1.4)
• GitHub Actions (tag-triggered) builds wheels/sdist and publishes to PyPI

Recommended (cargo-style) tool: cargo-release

cargo install cargo-release

# Patch release: bumps Cargo.toml version, commits (signed), creates signed tag, pushes
cargo release patch --execute

License

MIT — see LICENSE.