Skip to content

Latest commit

 

History

History
 
 

funcbench

funcbench

Benchmark and compare your Go code between commits or sub benchmarks. It automates the use of go test -bench to run the benchmarks and uses benchstat to compare them.

funcbench currently supports two modes, Local and GitHub. Running it in the Github mode also allows it to accept a pull request number and a branch/commit to compare against, which makes it suitable for automated tests.

Environment variables

  • GITHUB_TOKEN: Access token to post benchmarks results to respective PR.

Usage Examples

Clean git state is required.

usage: funcbench [<flags>] <target> [<bench-func-regex>] [<packagepath>]

Benchmark and compare your Go code between sub benchmarks or commits.
  - For BenchmarkFuncName, compare current with master: ./funcbench -v master
    BenchmarkFuncName
  - For BenchmarkFunc.*, compare current with master: ./funcbench -v master
    BenchmarkFunc.*
  - For all benchmarks, compare current with devel: ./funcbench -v devel .* or
    ./funcbench -v devel
  - For BenchmarkFunc.*, compare current with 6d280 commit: ./funcbench -v 6d280
    BenchmarkFunc.*
  - For BenchmarkFunc.*, compare between sub-benchmarks of same benchmark on
    current commit: ./funcbench -v . BenchmarkFunc.*
  - For BenchmarkFuncName, compare pr#35 with master: ./funcbench --nocomment
    --github-pr="35" master BenchmarkFuncName

Flags:
  -h, --help                 Show context-sensitive help (also try --help-long
                             and --help-man).
  -v, --verbose              Verbose mode. Errors includes trace and commands
                             output are logged.
      --nocomment            Disable posting of comment using the GitHub API.
      --owner="prometheus"   A Github owner or organisation name.
      --repo="prometheus"    This is the repository name.
      --github-pr=GITHUB-PR  GitHub PR number to pull changes from and to post
                             benchmark results.
      --workspace="/tmp/funcbench"
                             Directory to clone GitHub PR.
      --result-cache="funcbench-results"
                             Directory to store benchmark results.
  -n, --user-test-name="default"
                             Name of the test to keep track of multiple
                             benchmarks
  -t, --bench-time=1s        Run enough iterations of each benchmark to take t,
                             specified as a time.Duration. The special syntax Nx
                             means to run the benchmark N times
  -d, --timeout=2h           Benchmark timeout specified in time.Duration
                             format, disabled if set to 0. If a test binary runs
                             longer than duration d, panic.
  -l, --perflock             Enable perflock (you must have perflock installed
                             to use this)

Args:
  <target>              Can be one of '.', tag name, branch name or commit
                        SHA of the branch to compare against. If set to '.',
                        branch/commit is the same as the current one;
                        funcbench will run once and try to compare between
                        2 sub-benchmarks. Errors out if there are no
                        sub-benchmarks.
  [<bench-func-regex>]  Function regex to use for benchmark.Supports RE2
                        regexp and is fully anchored, by default will run all
                        benchmarks.
  [<packagepath>]       Package to run benchmark against. Eg. ./tsdb, defaults
                        to ./...

Building Docker Image

docker build -t prominfra/funcbench:master .

Triggering with GitHub comments

The benchmark can be triggered by creating a comment in a PR which specifies a branch to compare. The results are then posted back to the PR as a comment. The Github Actions workflow for funcbench can be found here.

The syntax is: /funcbench <branch|tag|commit> <benchmark function regex>.

  • See used regex for comment here.
  • The <benchmark function regex> expects the Benchmark prefix. It is anchored and passed to go test command, so need to anchor it in the comment.
Command Explanation
/funcbench master BenchmarkQuery.* Compare all the benchmarks matching BenchmarkQuery.* for master vs the PR
/funcbench feature-branch or /funcbench tag-name .* Compare all the benchmarks on feature-branch/tag-name vs the PR
/funcbench master BenchmarkQuery.* ./tsdb Compare all the benchmarks matching BenchmarkQuery.* for master vs the PR in package ./tsdb
/funcbench master Benchmark(?:Isolation.*|QuerierSelect) ./tsdb Compare all benchmarks matching Benchmark(?:Isolation.*|QuerierSelect) for master vs the PR

Notes:

  • Editing/Deleting a comment will not re-trigger the workflow of starting/stopping a benchmark. Only creating a comment starts a benchmark.
  • In case of funcbench, it automatically cleans up. So, no explicit stop command required.
  • Multiple comment lines are allowed:
/funcbench old_branch .*
The old_branch performs poorly, I bet mine are much better.