Skip to content

README.md

Latest commit

 

History

History
59 lines (42 loc) · 1.69 KB

README.md

File metadata and controls

59 lines (42 loc) · 1.69 KB

MicroAPI Benchmarks

These are local, reproducible benchmarks intended for README marketing and quick sanity checks.

They are not part of the normal unit/integration test suite because performance results are highly dependent on:

  • CPU / thermal throttling
  • OS / kernel
  • Python version
  • background processes
  • network stack details

Requirements

Install benchmark dependencies + required transports:

uv sync --extra bench --extra http --extra grpc

Run

uv run python -m benchmarks.run

You can tweak the run:

uv run python -m benchmarks.run --requests 10000 --warmup 500 --payload-bytes 256

Generate an SVG chart (for README)

uv run python -m benchmarks.run --requests 5000 --warmup 500 --payload-bytes 128 \
  --json-out benchmarks/latest.json \
  --svg-out benchmarks/benchmark.svg

What is measured

All benchmarks run on localhost and measure:

  • Req/s (throughput)
  • p50 / p95 latency

Compared setups:

  • FastAPI HTTP: uvicorn server + httpx client
  • MicroAPI HTTP: MicroAPI server + aiohttp transport + MicroAPI client
  • MicroAPI gRPC: MicroAPI server + custom HTTP/2 h2 transport + MicroAPI client
  • aiohttp raw baseline (optional): minimal aiohttp server/client without framework features

Interpreting results

  • FastAPI benchmark uses ORJSONResponse to avoid penalizing it with slow JSON serialization.
  • MicroAPI uses orjson internally for protocol envelopes and Pydantic for schemas.
  • MicroAPI gRPC and FastAPI HTTP are different protocols. The point is to show a realistic migration path: many teams build internal RPC over FastAPI/HTTP first, then switch to a dedicated RPC transport.