🏁 tenzir-bench

tenzir-bench measures the performance of Tenzir pipelines. It runs repeatable benchmarks against local binaries or container images and records results for comparison across builds and releases.

✨ Highlights

• ⏱️ Repeatable measurements: Run warmups and multiple measurement passes with consistent datasets and execution settings.
• 🧩 Reusable fixtures: Provision services such as Kafka or a Tenzir node and seed them with benchmark inputs.
• 🏎️ Flexible runners: Measure wall-clock time, CPU use, peak memory, and hardware counters with pluggable runners.
• 🔬 Build comparisons: Compare local binaries and Docker images directly, or evaluate runs against published release results.

📦 Installation

tenzir-bench requires Python 3.11 or newer. Run the latest compatible release from PyPI with uvx:

uvx tenzir-bench --help

uvx downloads the release into an isolated environment and caches subsequent invocations.

🚀 Quick start

Prepare the managed reference datasets:

uvx tenzir-bench prepare

Run the examples against a local Tenzir build:

uvx tenzir-bench run --tenzir ./build/bin/tenzir --benchmark examples/benchmarks

Compare two builds:

uvx tenzir-bench compare \
  --base ./build/baseline/bin/tenzir \
  --candidate ./build/candidate/bin/tenzir \
  examples/benchmarks \
  --compact

Build targets can also use the docker:// prefix, for example, docker://ghcr.io/tenzir/tenzir:main.

📚 Documentation

The documentation currently lives in this repository:

• Read the command guide to run, compare, synchronize, evaluate, and publish benchmarks.
• Read the benchmark authoring guide to define benchmarks, implementations, inputs, and fixtures.
• Explore the runnable examples for complete benchmark layouts.

📜 License

tenzir-bench is available under the Apache License, Version 2.0. See LICENSE for details.