gobenchdata is a tool for parsing and inspecting go test -bench data, and a GitHub Action for continuous benchmarking. It was inspired by the deno.land continuous benchmarks, which aims to display performance improvements and regressions on a continuous basis.

• GitHub Action
  • Setup
  • Configuration
    • inputs
      • Publishing
      • Checks
    • env
  • Pull Request Checks
  • Visualisation
• Command Line Interface
• Contributions

gobenchdata can be used as GitHub Action for uploading Go benchmark data as JSON to gh-pages and visualizing it with a generated web app or your own web application.

The default action, uses: bobheadxi/gobenchdata@v1, is distributed as a Docker container action. This means that the Go version benchmarks are run with are tied to the version of Go that ships with the gobenchdata action. For greater flexibility, such as to use custom Go versions, see custom setup.

Learn more about GitHub Actions in the official documentation.

For example, in .github/workflows/push.yml, using the new YAML syntax for workflows, a simple benchmark run and publish workflow would look like:

name: gobenchdata publish
on: push
jobs:
  publish:
    runs-on: ubuntu-latest
    steps:
    - name: checkout
      uses: actions/checkout@v2
    - name: gobenchdata publish
      uses: bobheadxi/gobenchdata@v1
      with:
        PRUNE_COUNT: 30
        GO_TEST_FLAGS: -cpu 1,2
        PUBLISH: true
        PUBLISH_BRANCH: gh-pages
      env:
        GITHUB_TOKEN: ${{ secrets.ACCESS_TOKEN }}

For greater flexibility and the ability to target custom Go versions, you can either:

• directly use the entrypoint.sh script, or some variant of it
• run gobenchdata action, which replicates the behaviour of the GitHub Action

Using gobenchdata action to replicate the Docker container action would look like:

name: gobenchdata publish
on: push
jobs:
  publish:
    runs-on: ubuntu-latest
    steps:
    - name: checkout
      uses: actions/checkout@v2
    - uses: actions/setup-go@v2
      with: { go-version: "1.18" }
    - name: gobenchdata publish
      run: go run go.bobheadxi.dev/gobenchdata@v1 action
      env:
        GITHUB_TOKEN: ${{ secrets.ACCESS_TOKEN }}
        # Instead of 'with: ...', provide environment variables and
        # prefix each input variable with 'INPUT'
        INPUT_PRUNE_COUNT: 30
        INPUT_GO_TEST_FLAGS: -cpu 1,2
        INPUT_PUBLISH: true
        INPUT_PUBLISH_BRANCH: gh-pages

Input variables are configured using jobs.<job_id>.steps.with.

The following inputs enable publishing - this merges and publishes benchmark results to a repository and branch of your choice. This is most useful in conjunction with the gobenchdata web application.

The following inputs are for enabling Pull Request Checks, which allow you to watch for performance regressions in your pull requests.

Environment variables are configured using jobs.<job_id>.steps.env.

Note that for GITHUB_TOKEN, it seems that pushes to gh-pages made by the default secrets.GITHUB_TOKEN might not trigger page builds. This issue can be resolved by using a personal access token instead.

Instead of publishing results, benchmark output can be used to pass and fail pull requests using CHECKS: true. To get started, set up the checks configuration:

go install go.bobheadxi.dev/gobenchdata@latest
gobenchdata checks generate

This will generate a file, gobenchdata-checks.yml, where you can configure what checks are executed. The checks are run against any benchmarks that match given package and benchmarks values, which should be provided as regular expressions.

checks:
- name: My Check
  description: |-
    Define a check here - in this example, we caculate % difference for NsPerOp in the diff function.
    diff is a function where you receive two parameters, current and base, and in general this function
    should return a negative value for an improvement and a positive value for a regression.
  package: .
  benchmarks: []
  diff: (current.NsPerOp - base.NsPerOp) / base.NsPerOp * 100
  thresholds:
    max: 10

The gobenchdata GitHub action eventually generates a JSON file with past benchmarks. You can visualize these continuous benchmarks by creating a web app that reads from the JSON benchmarks file, or by using gobenchdata. An easy way to get started is:

go install go.bobheadxi.dev/gobenchdata@latest
gobenchdata web generate --web.config-only .
gobenchdata web serve # opens visualization in browser

You can configure the web application using gobenchdata-web.yml. The configuration allows you to define groups of charts, where each group can be used to compare a set of benchmarks. Benchmarks are selected with regular expressions by package and benchmark names provided in the configuration.

Note that in each set of compared benchmarks, every metric will get its own chart. You can select which metrics to display using the metrics option.

title: gobenchdata web
description: Benchmarks generated using 'gobenchdata'
repository: https://github.com/bobheadxi/gobenchdata
benchmarksFile: benchmarks.json
chartGroups:
  - name: Demo Benchmarks
    description: |
      This is a demo for gobenchdata, a tool and GitHub action for setting up simple continuous
      benchmarks to monitor performance improvements and regressions in your Golang benchmarks!
    charts:
      - name: specify charts by package
        package: go.bobheadxi.dev\/gobenchdata\/demo
      - name: match on specific benchmarks across packages with glob patterns
        benchmarks: [ 'BenchmarkFib.' ]
  - name: More Demo Benchmarks
    description: Create multiple groups of benchmarks
    charts:
      - name: match by a combination of package and benchmarks
        package: go.bobheadxi.dev\/gobenchdata\/.
        benchmarks: [ 'BenchmarkPizzas.', '.FibSlow.' ]

You can output the entire web application (to publish to Github pages, for example) using:

gobenchdata web generate ./app

gobenchdata, which the GitHub Action leverages to manage benchmark data, is also available as a CLI:

go install go.bobheadxi.dev/gobenchdata@latest
gobenchdata help

The easiest way to use the CLI is by piping the output of go test -bench to it - gobenchdata will consume the output and generate a JSON report for you.

go test -bench . -benchmem ./... | gobenchdata --json bench.json

You can use this report to create your own charts, or just use the built-in web application:

gobenchdata web serve

gobenchdata can also execute checks for you to help you ensure performance regressions don't happen:

gobenchdata checks generate
gobenchdata checks eval ${base benchmarks} ${current benchmarks}

For more details on how to use checks, see the pull request checks documentation.

The CLI also embeds the same script that the GitHub Action offers - see custom setup for more details.

Please report bugs and requests in the repository issues!

See CONTRIBUTING.md for more detailed development documentation.