Evidence status
This site publishes no throughput multiplier or latency guarantee. The source contains a benchmark design, but that alone cannot substantiate a comparison. The illustration plates contain no runtime measurements.
Keep the workload accountable
- Producer latency: time the logging call, including formatting, filters, and validation.
- Delivery latency: observe the output, not merely enqueue completion.
- Saturation: record queue capacities, policies, accepted calls, delivered records, drops, and warnings.
- Shutdown: wait for every output boundary and report failed flushes.
Record Python and Rust versions, dependency versions, CPU, operating system, handler transport, message sizes, producer counts, warm-up, and repeat count. Compare equivalent formatting and destinations.
Start from the benchmark design
Read the source benchmark methodology. Performance results belong here when runnable artefacts, output accounting, and workload metadata exist. A smaller queue or discarded output is not an optimization result.