HaloLog and the Problem With a 23.9 ns Benchmark
How HaloLog reached 23.9 ns/op, why a Zap maintainer’s review changed the harness, and how zero-allocation claims became executable tests.
The article discusses the limitations of a benchmark score of 23.9 nanoseconds per operation for logging libraries. While the number is precise and can be repeated, it is incomplete and can be misleading without additional context such as the work unit, output contract, clock semantics, sink, concurrency model, hardware, toolchain, and competing configurations.
The benchmark measures the time taken to encode and dispatch a complete JSON record to io.Discard, but it does not account for the full ingestion process, which includes disk, network, batching, backpressure, or production log collector. The article emphasizes the importance of defining the operation precisely and measuring only the relevant aspects of logging performance.
It also highlights the need to distinguish between the idiomatic public hot path and other potential patterns. The author conducted an exploratory test on Windows, where changing the time encoder in Zap led to different results. The corrected test suite will report two named configurations: Zap production-default timestamp and Zap benchmark timestamp, to ensure transparency and fairness in the comparison.
Written by urgent.news from HackerNoon's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
