KubeElasti

Imported from KubeElasti/KubeElasti (AGENTS.md). Install upstream with npx skills add KubeElasti/KubeElasti. Copyright stays with the author.

AGENTS.md

Guidance for AI agents working in the KubeElasti repository.

What this is

KubeElasti is a Kubernetes-native controller that provides scale-to-zero: it scales a service to zero when there is no traffic and back up to one when traffic arrives. Two components:

• operator (operator/) - the controller/manager. Watches the ElastiService CRD, manages scaling, and switches services between "proxy" (scaled to zero, resolver in front) and "serve" (running) modes.
• resolver (resolver/) - the request proxy. Holds incoming requests while a scaled-to-zero target spins up, then forwards them; signals the operator to scale from zero.
• pkg (pkg/) - shared libraries used by both (k8shelper, scaling, logger, values, etc).

These are three separate Go modules stitched together by a go.work (Go 1.26.5).

Build, test, lint

Run from the repo root:

• make test - all tests (operator + resolver + pkg). Per-component: make test-operator, make test-resolver, make test-pkg.
• Per module you can also cd operator && make <target>: build, lint, lint-fix, fmt, vet, manifests, generate, test-e2e.
• Operator test/build targets auto-run manifests generate fmt vet first, so CRD manifests and deepcopy code stay in sync. If you change the API types in operator/api/, re-run cd operator && make manifests generate and commit the results.
• Lint is golangci-lint (config: .golangci.yaml).
• E2E tests live in tests/e2e; load tests in tests/load.

Code coverage: the goal is ~90% unit-test coverage across the codebase, reached incrementally. Every change you make - new code or edits to existing code - should ship with unit tests covering the lines and branches it touches, keeping coverage at or above that bar.

Use Go at /Users/tf/.homebrew/bin/go (1.26.5) if go on PATH is older.

Conventions

• Handle every error explicitly; no silent _ = on error returns. Use Result/wrapped errors idiomatically.
• Match surrounding style; don't add comments unless the code is genuinely non-obvious.
• Every commit is signed off: git commit -s. Never add an agent as co-author or put any agent signature in commits or PRs.
• Work on a branch, never commit directly to main.
• Never hand-edit auto-generated files: charts/elasti/README.md (helm readme generator), docs/index.yaml (helm index), CRD manifests, and CHANGELOG.md release-automation output.

Helm chart

The chart is charts/elasti/. Chart.yaml holds version/appVersion; values.yaml holds the operator and resolver image tags. Images and chart publish to ghcr.io/kubeelasti.

Releases

The release process is documented in RELEASE.md and automated by the release skill (.claude/skills/release/). In short: bump charts/elasti/Chart.yaml + charts/elasti/values.yaml

• CHANGELOG.md, open a PR to main, then create a v* GitHub release/tag which triggers .github/workflows/release.yaml to build, cosign-sign, SBOM, and publish images and the chart. The version in files carries no leading v (e.g. 0.1.31-rc1); the git tag does (v0.1.31-rc1).