Benchmarking HaloLog against zap, zerolog, slog, logrus, and phuslu/log: 23.9 ns/op

I’m sharing HaloLog, a structured logger I have developed for three years and run inside an interoperability project for the past two. Its goal is low-overhead structured logging without trading away JSON correctness, race safety, configured PII masking, or compile-checked telemetry fields. This is a public release and a request for independent reproduction of the measurements, not a claim about disk or network throughput.

The committed suite compares HaloLog with zap, zerolog, slog, logrus, and phuslu/log while encoding and dispatching complete JSON records to io.Discard. On the published Go 1.27 linux/amd64 host, HaloLog measured 23.9 ns/op for one complete record and 0 B/op, 0 allocs/op across all its measured scenarios. The repository documents the hardware, configurations, methodology, known differences, and correctness guardrails. These are results for those workloads and that host, not a universal performance claim.

If you reproduce or falsify a result, please include the Go version, hardware, command, and raw output. Useful counterexamples include an allocation, invalid JSON, a race, an unfair competitor configuration, or a workload in which HaloLog performs worse.