Package benchmarks contains benchmarks for slog.
These benchmarks are loosely based on github.com/uber-go/zap/benchmarks. They have the following desirable properties:
They test a complete log event, from the user's call to its return.
The benchmarked code is run concurrently in multiple goroutines, to better simulate a real server (the most common environment for structured logs).
Some handlers are optimistic versions of real handlers, doing real-world tasks as fast as possible (and sometimes faster, in that an implementation may not be concurrency-safe). This gives us an upper bound on handler performance, so we can evaluate the (handler-independent) core activity of the package in an end-to-end context without concern that a slow handler implementation is skewing the results.
We also test the built-in handlers, for comparison.