Acceptance of v0.1

This page checks v0.1 against the criteria of the framing letter of the project (November 2026): its objectives, its use cases, the scope of the MVP, its rules of good practice, and the gate of the release. Each criterion is marked:

• ✅ met, with the proof;
• ⚠️ met differently, with the reason;
• ⏳ not met yet, with the issue that follows it.

Release checked: v0.1.0-rc.3 (commit 4192480), the release that v0.1.0 is cut from. Date: 2026-11-10.

How to run it again

hack/acceptance.sh v0.1.0-rc.3

The script creates a fresh kind cluster (Kubernetes 1.37), installs the release from its install.yaml, downloads and checks vesperctl, then runs each check below and prints what it sees. It uses a kubeconfig of its own and deletes the cluster at the end. It needs kind, kubectl, curl, jq, Docker and the GitHub CLI. The public address over HTTPS needs a real server, and follows Try Vesper on a server.

Objectives

Objective	Indicator	Result
Create and destroy an environment from a PreviewEnvironment	full cycle shown on a local kind cluster	✅ the run creates, serves, updates, extends, expires and deletes previews
No orphan environment	0 Vesper namespace left after the TTL, checked by an end-to-end test	✅ the e2e test TestCLIExpiry checks that the namespace is gone; the run ends with 0 namespaces labelled app.kubernetes.io/managed-by=vesper
An idiomatic, tested Go controller	unit and envtest tests green in CI on every pull request	✅ the test job runs both with -race on every pull request; main is green
Understand each concept before using it	a learning journal, and an ADR per structural choice	⚠️ six ADRs (docs/adr); the journal has its template only: ⏳ to write
A project others can join	contribution guide, 10 good first issue, a first external contributor	✅ CONTRIBUTING.md, 10 open good first issues; ⏳ the first external contributions, #150 and #152, are in review and will land after v0.1.0
Prepare for the CNCF Sandbox	Apache 2.0 license and a neutral GitHub organization	✅ LICENSE, SPDX headers checked in CI, organization eventide-oss

Use cases

Use case	Result
1. Create an environment from a container image (by hand; from a pull request in v0.2)	✅ vesperctl create demo --image nginxinc/nginx-unprivileged:1.27-alpine becomes Ready and serves its page
2. Delete the environment when its TTL passes (when the pull request closes in v0.2)	✅ a preview with a TTL of 1 minute is deleted with its namespace
3. Extend or stop an environment early with vesperctl	✅ vesperctl extend demo --by 2h; vesperctl delete other --yes deletes it and its namespace
4. See the state, the events and the URL with vesperctl list and vesperctl get	✅ get shows the phase, URL, expiry, conditions and, since #151, the events
5. Enforce resource limits per environment	✅ a ResourceQuota and a LimitRange in each preview namespace; a pod over the quota is refused

Scope of the MVP

Item	Result
CRD PreviewEnvironment (v1alpha1): image, port, TTL, resource limits	⚠️ image, port, replicas, env and TTL are fields; resource limits are set by the administrator with operator flags (--quota-*, --default-*), not per preview, so that a user cannot raise their own budget. Limits per team are planned with the policies of v0.4.
Controller: a namespace per environment, with a Deployment, a Service, HTTP access and a ResourceQuota	✅ plus a LimitRange and a NetworkPolicy; HTTP access through an Ingress with --preview-domain
Guaranteed deletion: finalizer, owner references, TTL expiry	⚠️ finalizer and TTL ✅; the objects of a preview are tracked by labels instead of owner references, because an owner reference cannot point to another namespace (ADR-0002); deleting the preview namespace deletes them
Readable status: conditions and Kubernetes events	✅ conditions Ready, Available, Expired with reasons, and events at each change
CLI vesperctl: create, list, get, extend, delete	✅ all five, with table, JSON and YAML output and documented exit codes (1 for a missing preview, 2 for a usage error)
Basic metrics exposed by the controller	✅ vesper_previews_active per phase and vesper_previews_expired_total; 401 without a token
Unit tests, envtest and an end-to-end test on kind	✅ the test, integration and e2e jobs of CI

Good practices

Rule	Result
gofmt, go vet and golangci-lint in CI	✅ the vet and lint jobs, with gosec
Errors wrapped with %w, context.Context first	✅ followed in the code and checked in review
Table-driven tests run with -race	✅ make test
Idempotent reconciliation	✅ an object that matches is not written again (architecture)
Owner references on created objects	⚠️ labels instead, as above
A finalizer for cleanup	✅ vesper.eventide.dev/cleanup
Status in status.conditions, and events	✅
Versioned API, validated in the CRD schema	✅ v1alpha1; the image must be pinned, the TTL is bounded to 1m–168h, replicas to 1–5
Minimal RBAC, never cluster-admin	✅ operator permissions; no access to Secrets
A namespace per environment, with quotas and the restricted Pod Security level, and the weak isolation documented	✅ a privileged pod is refused; security model
Three test levels	✅
SemVer, Conventional Commits, changelog, GoReleaser releases	✅ Releasing
CONTRIBUTING, CODE_OF_CONDUCT, SECURITY, GOVERNANCE, issue templates	✅
Apache 2.0 and DCO	✅ the dco job checks every commit
An ADR per structural choice	✅ six ADRs
Aim for the OpenSSF Best Practices badge, in time	⏳ after v0.1, as the letter says
A learning journal	⏳ to write, as above

Gate: v0.1 published

Item	Result
README	✅ status, quickstart, links to the guides
License	✅
Contribution guide	✅
Repository visibility	✅ public, under the organization eventide-oss; the image on GHCR is public too

The run, in detail

On a fresh kind cluster, with v0.1.0-rc.3:

Check	Seen
Install	kubectl apply -f install.yaml rolls the operator out; it logs Starting Vesper {"version": "v0.1.0-rc.3"}; its image is pinned by digest
Release artifacts	the vesperctl archive passes its checksum and its attestation; the image passes its attestation, for linux/amd64 and linux/arm64, user 65532:65532
Create	Ready; the namespace holds the Deployment, Service, ResourceQuota, LimitRange and NetworkPolicy; the page answers through a port-forward
Validation	latest, a TTL of 30s and 6 replicas are refused by the API server, exit code 2, with a message naming the field
Pod Security	a privileged pod is refused: violates PodSecurity "restricted:latest"
Quota	a pod asking for 5Gi is refused: exceeded quota: preview
Network	from a preview: another preview blocked, the API server blocked, the internet reached
Update	a new image rolls out, the preview is Ready again
Extend	extended by 2h0m0s: ttl 3h0m0s
Expiry	a 1-minute preview and its namespace are gone
Delete	the preview and its namespace are gone
Status	vesperctl get shows the conditions and the events (NamespaceCreated, Provisioning, DeploymentAvailable)
CLI	list and get in table, JSON and YAML; exit 1 for a missing preview, 2 for an unknown output format
Metrics	401 without a token; with one, every phase and vesper_previews_expired_total 1 after the expiry
Cleanup	0 Vesper namespaces left

On a real server (#96, with v0.1.0-rc.2, which differs from rc.3 by the events of vesperctl get and docs): k3s on a VPS, a preview opened from a browser and a phone at its public address, over HTTPS with a Let’s Encrypt wildcard certificate, and no Secret in the preview namespace. That run found an isolation gap, fixed in rc.2 (GHSA-m437-fj8p-6x8h).

Follow-ups

Issue	What
#150, #152	the first external contributions, in review; they land after v0.1.0
#155	the message of Provisioning during a rollout reads 1/1 replicas available, seen in this run
#154	TestList can fail when it runs across a second
—	the learning journal (docs/journal)
—	the OpenSSF Best Practices badge, and checking the name before any CNCF application
P5	previews from pull requests (GitHub App), private images, protected URLs

Verdict: v0.1 meets the MVP of the framing letter. The two ⚠️ items are choices recorded above; the ⏳ items do not block the release.