Performance

These are producer-side numbers: what a LOG_* call costs the thread that makes it.

Benchmark Highlights

Recorded on an AMD Ryzen 9 5900HX, Windows 11, MSVC 19.44 /O2, with log output on a local NVMe SSD (2026-10-22):

Producer-side measurement	Result
Literal message, 1 producer	14.35 million calls/sec
Literal message, 8 producers	18.91 million calls/sec
Three-argument message, 8 producers	7.48 million calls/sec
Mean call latency, literal message	153 ns

With 4 producer processes of 2 threads each, a call took about 185-200 ns with thread rings in shared memory, against 800-950 ns through the shared queue (multi-process results).

Call timings exclude the writer's drain and flush time. A burst a thread ring cannot hold spills to the shared rings, which can overwrite their oldest entries when overloaded. The benchmark README has the full results, comparisons, and methodology, plus build/run instructions for reproducing the numbers on your own hardware.