tree afde757983423b68455d0df3bde6fe592d56b219
parent a68e582fa157abff310e2a9a287b21a1fe1e1629
author Jonathan Amsterdam <jba@google.com> 1673631799 -0500
committer Jonathan Amsterdam <jba@google.com> 1674570587 +0000

slog: re-run slog, zerolog, zap benchmarks

slog is as fast as zap, as expected (output from benchstat):

Attrs/fastText_discard/5_args-8      518ns ± 3%     486ns ± 7%      ~     (p=0.056 n=5+5)
Attrs/fastText_discard/10_args-8     919ns ±13%     918ns ± 5%      ~     (p=0.841 n=5+5)

slog is slower than zerolog:

Attrs/fastText_discard/5_args-8      183ns ± 4%     486ns ± 7%  +165.39%  (p=0.008 n=5+5)
Attrs/fastText_discard/10_args-8     283ns ± 4%     918ns ± 5%  +224.63%  (p=0.008 n=5+5)

These numbers understate zerolog's advantage, because it produces JSON
and the slog handler used for benchmarking just concatenates text.
With the slog JSONHandler, the times are 925ns for 5 args and 1630ns
for 10 args.

Most of zerolog's advtange is likely due to its design, which trades
flexibility and safety for speed. Also, zerolog doesn't lock, but that
seems to add only about 200ms to the time. There are no obvious
improvements to be made in the slog JSON implementation, but we will
try to improve it by profiling.

It's important to remember that these benchmarks discard the output,
neglecting I/O latency, which can be significant.

Change-Id: Ie408873465055892e9b70a98aff5c1568ee4175b
Reviewed-on: https://go-review.googlesource.com/c/exp/+/461998
Reviewed-by: Alan Donovan <adonovan@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Jonathan Amsterdam <jba@google.com>