search

gohugoio / hugo

The world’s fastest framework for building websites.
https://gohugo.io
Apache License 2.0
73.49k stars 7.37k forks source link

tests: Refine test cases for BenchmarkTaxonomiesGetTerms #12612

Closed razonyang closed 1 week ago

razonyang commented 1 week ago

The tag terms on PR #12611 seems too small ([a,b,c,d,e,f]), this PR will create a large number of tags to cover more situations. However I just modified the test cases, since I don't have much ability/knowledge to improve it.

The result of benchmark on my laptop as follows.

Running tool: /usr/bin/go test -benchmem -run=^$ -bench ^BenchmarkTaxonomiesGetTerms$ github.com/gohugoio/hugo/hugolib

goos: linux
goarch: amd64
pkg: github.com/gohugoio/hugo/hugolib
cpu: 11th Gen Intel(R) Core(TM) i7-11800H @ 2.30GHz
BenchmarkTaxonomiesGetTerms/pages_100-16                 100      10308688 ns/op     6421780 B/op      73897 allocs/op
BenchmarkTaxonomiesGetTerms/pages_1000-16                  9     132214460 ns/op    49648174 B/op     556238 allocs/op
BenchmarkTaxonomiesGetTerms/pages_10000-16                 1    7714456586 ns/op    476423784 B/op   5419983 allocs/op
BenchmarkTaxonomiesGetTerms/pages_20000-16                 1    41908415387 ns/op   951093296 B/op  10836903 allocs/op
PASS
ok      github.com/gohugoio/hugo/hugolib    55.180s

Please close it if I'm understand it wrongly.

jmooring commented 1 week ago

The number of generated tags seems unrealistic, and the existing test covers the underlying problem.

razonyang commented 1 week ago

the existing test covers the underlying problem.

It seems doesn't cover the case of comparing different number of taxonomy terms.

razonyang commented 1 week ago

If I correctly understand the benchmark above, the speed seems getting slower with large number of terms, maybe it should also be considered a improvable case.

Please close it if I'm wrong, but I see some users to generate a lot of terms on their site.

bep commented 1 week ago

Thanks for this, but the original benchmarks are good enough. The problem with the original GetTerms code was obvious, and the benchmarks confirmed the theory.

bep commented 1 week ago

As to running Go benchmarks, I use this tool:

https://github.com/bep/gobench

... lacks some documentation, but is very useful when comparing branches etc. as it supports benchstat and pprof under the hood.