tenacity-rs vs tenacity Benchmarks

This directory contains a benchmark suite comparing tenacity-rs (Rust core) with the original tenacity (pure Python).

Installation

From the project root (not from benchmarks/):

# Install tenacity-rs in development mode
pip install -e .

# Install tenacity for comparison
pip install tenacity

Or install only tenacity-rs and run with --no-tenacity to benchmark tenacity-rs alone.

From the project root:

cd benchmarks
python bench.py

Or from anywhere:

python benchmarks/bench.py

--iterations N — Number of timed iterations per benchmark (default: 500).
--warmup N — Warmup iterations before timing (default: 20).
--no-tenacity — Skip tenacity; only run tenacity-rs (useful if tenacity is not installed).
--json FILE — Save raw results (mean, speedup, etc.) to a JSON file.
--markdown FILE — Save a Markdown table of results.

Example:

python bench.py --iterations 1000 --warmup 30 --json results.json --markdown results.md

Speedup — tenacity_time / tenacity_rs_time. Values > 1 mean tenacity-rs is faster.
Mean ± stdev — Average duration and standard deviation across iterations.
Legend:
- 🚀 — tenacity-rs is more than 1.5× faster
- ✓ — tenacity-rs is faster
- ⚠️ — tenacity-rs is slower (e.g. when dominated by Python callbacks or FFI)

Retry logic (stop/wait/retry checks) — tenacity-rs is typically 2–5× faster due to the Rust core.
Scenarios with real sleep() — Both are similar; most time is spent in the OS sleep.
Heavy use of Python callbacks (before/after/before_sleep) — tenacity-rs can be slightly slower due to FFI when calling back into Python.
High retry counts — tenacity-rs tends to shine because the per-attempt overhead is lower.

Close other heavy applications to reduce variance.
Run multiple times and compare; use --iterations 1000 for smoother means.
On laptops, plug in and avoid thermal throttling; or run in a cool environment.
Results vary by machine and Python version; use the suite to compare relative performance on your setup.

JSON — Machine-readable: per-benchmark mean times and speedup for plotting or further analysis.
Markdown — Table suitable for pasting into docs or README.