Awesome
<img src="img/logo.png" width="350">
Abstract
cob compares benchmarks between the latest commit (HEAD) and the previous commit (HEAD{@1}). The program will fail if the change in score is worse than the threshold. This tools is suitable for CI/CD to detect a regression of a performance automatically.
cob runs go test -bench before and after commit internally, so it depends on go command.
CAUTION: Note that git reset is executed when you run cob. You should commit all changes before running cob.
Table of Contents
<!-- TOC -->Continuous Integration (CI)
See cob-example for details.
GitHub Actions
name: Bench
on: [push, pull_request]
jobs:
test:
name: Bench
runs-on: ubuntu-latest
steps:
- name: Set up Go 1.13
uses: actions/setup-go@v1
with:
go-version: 1.13
id: go
- name: Check out code into the Go module directory
uses: actions/checkout@v1
- name: Install cob
run: curl -sfL https://raw.githubusercontent.com/knqyf263/cob/master/install.sh | sudo sh -s -- -b /usr/local/bin
- name: Run Benchmark
run: cob
Travis CI
dist: bionic
language: go
go:
- 1.13.x
before_script:
- curl -sfL https://raw.githubusercontent.com/knqyf263/cob/master/install.sh | sudo sh -s -- -b /usr/local/bin
script:
- cob
CircleCI
version: 2
jobs:
bench:
docker:
- image: circleci/golang:1.13
steps:
- checkout
- run:
name: Install cob
command: curl -sfL https://raw.githubusercontent.com/knqyf263/cob/master/install.sh | sudo sh -s -- -b /usr/local/bin
- run:
name: Run cob
command: cob
workflows:
version: 2
build-workflow:
jobs:
- bench
Example
Override a command to measure benchmarks
To measure benchmarks by make bench, you can use -bench-cmd and -bench-args options.
$ cob -bench-cmd make -bench-args bench
Run only those benchmarks matching a regular expression
$ cob -bench-args "test -bench Append -benchmem ./..."
2020/01/12 17:32:30 Run Benchmark: 4363944cbed3da7a8245cbcdc8d8240b8976eb24 HEAD{@1}
2020/01/12 17:32:32 Run Benchmark: 599a5523729d4d99a331b9d3f71dde9e1e6daef0 HEAD
Result
======
+-----------------------------+----------+---------------+-------------------+
| Name | Commit | NsPerOp | AllocedBytesPerOp |
+-----------------------------+----------+---------------+-------------------+
| BenchmarkAppend_Allocate-16 | HEAD | 179.00 ns/op | 117 B/op |
+ +----------+---------------+-------------------+
| | HEAD@{1} | 115.00 ns/op | 23 B/op |
+-----------------------------+----------+---------------+-------------------+
Comparison
==========
+-----------------------------+---------+-------------------+
| Name | NsPerOp | AllocedBytesPerOp |
+-----------------------------+---------+-------------------+
| BenchmarkAppend_Allocate-16 | 55.65% | 408.70% |
+-----------------------------+---------+-------------------+
Show only benchmarks with worse score
$ cob -only-degression
2020/01/12 17:48:35 Run Benchmark: 4363944cbed3da7a8245cbcdc8d8240b8976eb24 HEAD{@1}
2020/01/12 17:48:38 Run Benchmark: 599a5523729d4d99a331b9d3f71dde9e1e6daef0 HEAD
Comparison
==========
+-----------------------------+---------+-------------------+
| Name | NsPerOp | AllocedBytesPerOp |
+-----------------------------+---------+-------------------+
| BenchmarkAppend_Allocate-16 | 52.34% | 347.83% |
+-----------------------------+---------+-------------------+
2020/01/12 17:48:39 This commit makes benchmarks worse
Specify a threshold
The following option means the program fails if a benchmark score gets worse than 50%.
$ cob -threshold 0.5 ./...
Specify a base commit compared with HEAD
By default, cob uses HEAD~1. If you compare benchmarks with different commit, you can use --base option.
$ cob --base origin/master ./...
Compare only memory allocation
You can use -compare option.
$ cob -compare B/op
2020/01/15 14:46:31 Run Benchmark: 4363944cbed3da7a8245cbcdc8d8240b8976eb24 HEAD~1
2020/01/15 14:46:33 Run Benchmark: 599a5523729d4d99a331b9d3f71dde9e1e6daef0 HEAD
Result
======
+-----------------------------+----------+---------------+-------------------+
| Name | Commit | NsPerOp | AllocedBytesPerOp |
+-----------------------------+----------+---------------+-------------------+
| BenchmarkAppend_Allocate-16 | HEAD | 179.00 ns/op | 121 B/op |
+ +----------+---------------+-------------------+
| | HEAD@{1} | 104.00 ns/op | 23 B/op |
+-----------------------------+----------+---------------+-------------------+
| BenchmarkCall-16 | HEAD | 0.50 ns/op | 0 B/op |
+ +----------+---------------+ +
| | HEAD@{1} | 0.49 ns/op | |
+-----------------------------+----------+---------------+-------------------+
Comparison
==========
+-----------------------------+---------+-------------------+
| Name | NsPerOp | AllocedBytesPerOp |
+-----------------------------+---------+-------------------+
| BenchmarkAppend_Allocate-16 | - | 426.09% |
+-----------------------------+---------+-------------------+
| BenchmarkCall-16 | - | 0.00% |
+-----------------------------+---------+-------------------+
2020/01/15 14:46:35 This commit makes benchmarks worse
Skip running cob
If your commit message contains [skip cob], cob is skipped.
$ git add README.md
$ git commit -m "[skip cob] update README.md"
$ cob
2020/04/19 12:46:57 [skip cob] is detected, so the benchmark is skipped
Usage
NAME:
cob - Continuous Benchmark for Go project
USAGE:
cob [global options] command [command options] [arguments...]
COMMANDS:
help, h Shows a list of commands or help for one command
GLOBAL OPTIONS:
--only-degression Show only benchmarks with worse score (default: false)
--threshold value The program fails if the benchmark gets worse than the threshold (default: 0.2)
--base value Specify a base commit compared with HEAD (default: "HEAD~1")
--compare value Which score to compare (default: "ns/op,B/op")
--bench-cmd value Specify a command to measure benchmarks (default: "go")
--bench-args value Specify arguments passed to -cmd (default: "test -run '^$' -bench . -benchmem ./...")
--help, -h show help (default: false)
Q&A
Benchmarks with the same name
Specify a package name.
$ cob -bench-args "test -bench . -benchmem ./foo"
$ cob -bench-args "test -bench . -benchmem ./bar"
A result of benchmarks is unstable
You can specify -benchtime.
$ cob -bench-args "test -bench . -benchmem -benchtime 10s ./..."
License
This repository is available under the MIT
Author
Teppei Fukuda (knqyf263)