Add continuous benchmarking with CodSpeed

[codspeed-hq]

What

Sets up continuous performance measurement for the CLI with CodSpeed, and adds benchmarks for a number of hot paths that were not covered yet.

Changes

CI (.github/workflows/codspeed.yml)

• New codspeed workflow, running on pull requests, pushes to master/release branches, and workflow_dispatch (the latter allows CodSpeed to backfill initial data).
• Uses the walltime instrument, which is the only instrument currently supported for Go.
• Authentication uses OIDC (id-token: write), so no token secret is needed.
• Follows the conventions of the existing workflows: actions pinned by commit hash with a version comment, persist-credentials: false, explicit permissions, ubuntu-24.04 runner, Go 1.26.4 via actions/setup-go.
• Like the host job in test.yml, the workflow symlinks vendor.mod/vendor.sum to go.mod/go.sum so the benchmarks run in module mode against the vendored dependencies.
• Only the packages that contain benchmarks are passed to go test -bench, to keep the CI run short. The cli/command/formatter/tabwriter benchmarks are intentionally left out: they are a copy of the upstream text/tabwriter benchmarks and include cases with 100k rows, which take minutes on their own.

Benchmarks

Existing benchmarks (BenchmarkStatsFormat, BenchmarkPrettyPrintInfo) are picked up as-is. Added benchmarks for paths that are hit on virtually every CLI invocation:

Benchmark	Package	What it covers
BenchmarkAddCommands	cli/command/commands	Constructing the cobra command-tree, done on every invocation of the CLI and when generating completion scripts and docs
BenchmarkLoadFromReader	cli/config	Loading config.json
BenchmarkParseRun	cli/command/container	Parsing docker run flags into container-, host- and networking-config (minimal and "many flags" cases)
BenchmarkMountOptSet	opts	Parsing --mount values (volume, bind, tmpfs)
BenchmarkContainerWrite	cli/command/formatter	Rendering docker ps output (table, table with size, quiet, raw, json, custom format)
BenchmarkImageWrite	cli/command/formatter	Rendering docker image ls output (table, table with digest, quiet, raw, json, custom format)
BenchmarkParse / BenchmarkExecute	templates	Parsing and executing --format templates

The new benchmarks use b.Loop() and b.ReportAllocs(), and are placed in the existing _test.go files next to the tests of the code they measure.

Docs

• CodSpeed badge added to README.md.
• A short "Benchmarks" section added to TESTING.md describing the naming convention, how to run benchmarks locally, and how to reproduce the CodSpeed run.

Verification

The full set was run locally with the CodSpeed runner in walltime mode (codspeed run --skip-upload --mode walltime -- go test -bench=. ...): 26 benchmarks across the 7 packages were discovered and measured, in roughly two minutes. golangci-lint run (v2.10.1, the version used by make lint) and gofumpt are clean on the touched packages, and go test still passes for them.

Note on runners

docker/cli is hosted on a personal account here, and CodSpeed macro runners are only available for organization repositories, so the workflow uses the standard ubuntu-24.04 runner. Walltime measurements on shared GitHub-hosted runners have significantly higher variance than on macro runners, so expect noisier results and treat small deltas with care. See https://codspeed.io/docs/instruments/walltime/index#usage-on-personal-github-accounts.

Next steps

• Merge, then let the first run on master establish the performance baseline (or trigger the workflow manually to backfill data).
• If the run turns out to be too slow as more benchmarks are added, the go test -bench invocation can be sharded across matrix jobs with -bench=<filter>.
• If the repository is moved to an organization, switch the runner to codspeed-macro for much more stable walltime results.

[codspeed-hq]

Congrats! CodSpeed is installed 🎉

🆕 27 new benchmarks were detected.

You will start to see performance impacts in the reports once the benchmarks are run from your default branch.

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

Detected benchmarks

• BenchmarkAddCommands: 427.5 µs
• BenchmarkStatsFormat: 161.9 µs
• many_flags: 33.2 µs
• minimal: 17 µs
• custom: 666.4 µs
• json: 3 ms
• quiet: 100.5 µs
• raw: 1.7 ms
• table: 1.3 ms
• table-with-size: 1.5 ms
• custom: 1.8 ms
• json: 4 ms
• quiet: 1.2 ms
• raw: 2.1 ms
• table: 2.4 ms
• table-with-digest: 2.6 ms
• BenchmarkPrettyPrintInfo: 9.6 µs
• BenchmarkLoadFromReader: 13.9 µs
• bind: 801 ns
• tmpfs: 781 ns
• ...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

---