Repository navigation
feat(business-site): lift the staging ingress hold - #6214
Conversation
Turn on the business-site Tenant's own application and WebApp at the
already-verified v3.0.0 package. The platform-owned staging hold still
denies gateway traffic, so the site is not reachable yet.
- Admit exactly business-site.platform.devantler.tech for HTTPRoutes in
the business-site namespace; the catch-all denied every route there.
- Admit a tenant egress rule that is exactly {}: Cilium allows nothing
through it, and the package uses it to select egress default deny.
- Let the canonical source guard accept the source-only bootstrap shape
or both switches on together, never a half-enabled Tenant.
- Approve the moved Tenant ledger line and static-scan instance hash.
Part of #5247.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The ownership-label replay pins each tenant family's exact Kyverno counts. The business-site route fixtures add one admitted and four refused routes, and the empty-egress fixture adds one admitted policy. Part of #5247. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The business-site package now routes its public pages on devantler.tech and www, and its portal on portal.devantler.tech. The shared Gateway had no listener or certificate for those names, and the route allow-list admitted only the staging name. - A platform-applied Certificate in the business-site namespace covers all three names through the shared Cloudflare ClusterIssuer, like world-at-ruin's zone-trial certificate. - A ReferenceGrant lets the Gateway read only that Secret. - Three HTTPS listeners, named for the sections the package's routes attach to, terminate TLS with it. - restrict-tenant-route-hostnames admits the three names beside the staging name; fixtures cover both new routes, an unlisted subdomain and a *.devantler.tech wildcard. The apex and www keep their GitHub Pages records until the cutover, since external-dns never overwrites a record it does not own. Part of #6177 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… builds on its settled binding Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
v3.2.1 is the first business-site release with the two-host runtime and the package-owned public and portal routes. It is signed by the approved publish-app revision d8b5d0e9, as v3.0.0 was. - The binding and the Tenant digest are the unedited output of `bootstrap-tenant-source.sh --prepare-canonical-upgrade --tag v3.2.1`, run against the settled v3.0.0 binding from #6197 and a read-only production context. The preparation verified the release (cosign, tag v3.2.1, source 16cf60a0, run 43400872692) and the live v3.0.0 source (Ready, generation 3). - The Tenant keeps #6174's enabled flags; only its digest moves to sha256:eec71bcf…c66e4. - The rendered authorization surface and the static-scan baseline move only the business-site Tenant entry. Part of #6177 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The signed staging application now runs behind the hold, so gateway traffic may reach it. The hold's name stays reserved, so only the platform can put it back. The customer identity hold is unchanged. Part of #6177 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…the image check trusts v3.0.0 and v3.2.1 were signed by publish-app v19.0.12, which the tenant source approved but production's image admission check does not, so the staging Deployment was denied. v3.2.2 is signed by v19.0.5, which the admission check, the node pull rule and the tenant source all trust. Business-site's approvals become v19.0.12 (the binding it moves from) and v19.0.5 (the one it moves to); v8.0.0 is no longer used by it. The upgrade record, Tenant digest and publisher, scan-baseline hash and ledger line come from the platform's own preparation and validators. Part of #6177 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…t-staging-hold Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Part of #6177 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
✅
|
✅
|
The approvals library drops a consumer's first row once its publisher move lands, so the revision being moved from (v19.0.12) goes first and the destination (v19.0.5) second. Part of #6177 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> # Conflicts: # scripts/canonical-tenant-source-bindings.json
…t-staging-hold Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
devantler
left a comment
There was a problem hiding this comment.
🤖 Generated by the Agentic Engineer
Self-review (fallback — CodeRabbit, Codex and Cursor Bugbot unavailable)
Reviewed commit: 00b28fa
- CodeRabbit: rate limited on this repository at 2026-11-13T15:13:53Z (#6188, "Review limit reached", next included review in 52 minutes)
- Codex: usage limit at 2026-11-13T14:28:31Z (#6188, "You have reached your Codex usage limits for code reviews")
- Cursor Bugbot: no reaction, check run or comment on this repository 45 minutes after "bugbot run" on #6188 at 2026-11-13T14:28:15Z; last platform reply (#6151) was "GitHub account mismatch"
Verdict: 2 findings (P0: 0, P1: 0)
Taken out of the merge queue to honour this PR's own merge condition: it merges only after #6188 has deployed and the staging app is healthy behind the hold. At 17:05Z the staging app was not healthy. Its Kubernetes deployment was being refused by the image-signature policy (the business-site Kustomization reports a failed health check), and #6188, which moves the app to an admitted release, was still in the queue at position 4. With both queued back to back, this PR would have merged before #6188 could be checked live. Next step: once #6188 has merged and Flux has applied it, I check that the app is Ready and then put this PR back in the queue at the same head (00b28fa). |
Blocker: cleared 2026-11-14 02:36Z — platform#6243 merged and Flux applied business-site v3.2.3 (manifests |
…lift-staging-hold # Conflicts: # k8s/providers/hetzner/apps/business-site/tenant.yaml # scripts/canonical-tenant-source-bindings.json # scripts/publish-workflow-canonical-approvals.tsv # scripts/rgd-template-static-scan-baseline.tsv # scripts/validate-eks-ci-role-policy/approved-surface.txt
@coderabbitai review New head 6b6dfdc merges main (platform#6243, business-site v3.2.3) into this branch. Conflicts were resolved to main's values in the tenant source, bindings, approvals, static-scan baseline and approved surface; only the tenant comment keeps this PR's "hold is lifted" wording. |
Action performed
|
|
No actionable comments were generated in the recent review. Walkthrough
Merge Risk:
|

Why
The business-site staging application was installed behind an ingress hold until it ran healthy, and the portal and public pages can only be checked live once gateway traffic reaches it.
What
Removes the staging ingress hold and keeps its name reserved, so only the platform can put it back; the customer identity hold stays. It merges only after #6188 has deployed and the application is confirmed healthy behind the hold, and it is taken out of the queue otherwise.
Part of #6177