Skip to main content
SnowOps Labs

Decisions

ADR 0001 — Cut cloud runtimes and commercial surfaces

Architecture decision records — why it was built this way.

docs/adr/0001-cut-cloud-and-commercial-scope.md
On this page

Status: Accepted Date: 2026-07-26 Wave: W0

Context

v1 shipped AKS, EKS and GKE runtimes with Terraform modules, plus a full commercial layer — pack marketplace with OCI distribution, entitlement tiers, CE/Pro/Enterprise editions, and a certificate issuance framework — roughly 2,000 lines of Go plus CI workflows, registry tooling and docs.

None of the cloud runtimes were ever executed against a real cloud account. Tasks 038 and 039 ("verify AKS end to end", "verify EKS end to end") sat open from the first milestone to the last. Meanwhile the commercial layer served zero users, since the project has no users at all yet.

The combined effect was breadth at the expense of depth: the core loop that every user hits first — start a cluster, run a scenario, break it, fix it — had no cancellation, no durable state, and no shell or UI tests.

Decision

Delete both surfaces in v2.

  • Cloud: remove runtimes/{aks,eks,gke}, foundation/terraform, ACR/ECR build strategies, docs/cloud-runtimes.md and the cloud CI workflows. Local (k3d, kind) and incluster remain.
  • Commercial: remove pkg/pack, pkg/entitlement, pkg/edition, pkg/credential, the marketplace API and UI, registry/, packs/ and sdk/pack-template. Keep pkg/extension as a small documented seam.

Alternatives considered

Alternative Why rejected
Keep and harden everything ~35% of remaining effort spent on features with no users, while the core stays weak
Keep packs, cut monetisation Pack distribution solves a problem git and a content path already solve (see ADR-0008)
Keep cloud as unverified/experimental An untested code path in the repo is a liability that still needs maintenance, review and CI time

Consequences

  • Easier: the golden path gets all the attention; CI is fast because kind is fast; there is no cloud cost to test anything.
  • Harder: users wanting a managed-cluster demo must bring their own kubeconfig. We accept this — the product's value is the simulation, not the provisioning.
  • Reversible: the runtime profile contract is unchanged, so a cloud profile can be reintroduced later by anyone with an account to verify it against.
  • The removed v1 code and plan are preserved by the maintainers, outside the public tree.