Skip to content

fix: merge PR #518 into main — let cluster own Gateway, drop ListenerSet#520

Closed
TineoC wants to merge 2 commits into
mainfrom
fix/merge-main-518
Closed

fix: merge PR #518 into main — let cluster own Gateway, drop ListenerSet#520
TineoC wants to merge 2 commits into
mainfrom
fix/merge-main-518

Conversation

@TineoC

@TineoC TineoC commented Jul 14, 2026

Copy link
Copy Markdown
Collaborator

Merges PR #518 (commit 809babe) into main.

Per @themightychris's findings in cfp-live-cluster#160:

  • ListenerSet is silently ignored by Envoy Gateway v1.7.3 (XListenerSet CRD not found)
  • cluster-issuer is wrongletsencrypt-prod solves ACME over ingress-nginx requiring PROXY protocol
  • certificateRef points to wrong secret — should be <app>-gw-tls, not legacy Ingress balancer-tls

The fix: this repo owns only the HTTPRoute (paths, ports, backend services). The cluster owns listeners, hostnames, TLS, and issuers. The route attaches to a cluster-provided Gateway by name and inherits hostnames from its listeners.

Closes CodeForPhilly/cfp-live-cluster#160

themightychris and others added 2 commits July 14, 2026 14:32
This repo cannot know the cluster it lands in, and gateway-listeners.yaml was
trying to. Between them, the ListenerSet and its routes hardcoded four facts
about cfp-live-cluster — the shared Gateway's name and namespace, the
ClusterIssuer, the TLS Secret name, and the ListenerSet API itself — and three
were wrong there:

  - cert-manager.io/cluster-issuer: letsencrypt-prod is dead. It solves ACME
    over ingress-nginx, which requires PROXY protocol on every connection and
    is therefore unreachable from inside the cluster. No certificate has issued
    through it since May.
  - certificateRefs: balancer-tls is the legacy Ingress Secret. Gateway certs
    are <app>-gw-tls.
  - ListenerSet is not reconciled by Envoy Gateway v1.7.3. It watches
    XListenerSet and logs "XListenerSet CRD not found, skipping XListenerSet
    watch". The object applies cleanly and is then ignored — the app just has
    no listener, with nothing in any log to say why.

None of that is a mistake anyone could have avoided from this repo. It isn't
knowable here. So stop trying to know it.

Split on what each side actually knows:

  this repo -> HTTPRoute: paths, backend Services, ports. App facts. They
               change when the app changes, in the same commit.
  cluster   -> Gateway: listeners, hostnames, TLS, issuers. Cluster facts. They
               differ per environment and are shared with every other app.

The route now attaches to a Gateway the cluster provides, named after the app,
and declares no hostnames — a route that omits hostnames inherits them from the
listeners it attaches to. So one file serves balancerproject.org in live and
balancer.sandbox.k8s.phl.io in sandbox, with no patches and no placeholders.
The HOSTNAME_PLACEHOLDER substitutions and the overlay routing patches are gone.

Drops balancer-http-redirect too: the cluster already redirects every hostname
reaching it on port 80. Worse than redundant — a route claiming an exact
hostname on the shared HTTP listener outranks the catch-all redirect and can
shadow cert-manager's ACME solver route, which is what produced the apex 404
that #517 set out to fix. That 404 was a stuck certificate, and it cleared the
moment the cert issued.

Adds a README stating the contract, so the next person doesn't have to infer it.

Refs CodeForPhilly/cfp-live-cluster#166, #168
@themightychris

Copy link
Copy Markdown
Member

the release flow handles this

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants