Benchmarks¶
httpr's performance is tracked three ways. All of them run in GitHub Actions.
Comparison with other clients¶
Every Monday, benchmark/benchmark.py
runs httpr against the latest release of requests, httpx, httpx2, aiohttp,
curl_cffi, pycurl and tls_client: 400 requests per cell against a local uvicorn
server with gzip responses, in sync, async and multi-threaded modes. Wall-clock
seconds, lower is better.
The full tables and raw CSVs are published with the chart. Absolute numbers depend on the GitHub runner, so compare rows within a chart rather than charts across weeks.
httpr over time¶
On every push to main, tests/benchmark/test_performance.py runs under
pytest-benchmark and the results are appended to the
live trend chart.
The runners are shared, so a single run can swing by 1.5x; look at the trend,
not at individual points.
Timings per pull request¶
CodSpeed runs
tests/benchmark/codspeed/ under its walltime instrument on every pull request
and reports the difference against main on the PR. Walltime includes the
socket I/O that dominates an HTTP client, which CPU simulation leaves out. It
runs on a shared GitHub runner, so small differences can be noise.
Running locally¶
See benchmark/README.md.
