fix: the EnvoyProxy must state its replica count, or the data plane stays down

charts/eduide-cluster/README.md

@@ -22,6 +22,7 @@ cert-manager issuers. Install once per cluster, before any eduide release.
 | envoyProxy.labels | object | `{}` |  |
 | envoyProxy.name | string | `"theia-shared-gateway"` |  |
 | envoyProxy.namespace | string | `"envoy-gateway-system"` |  |
+| envoyProxy.replicas | int | `1` | Data plane replica count, rendered into the spec at `provider.kubernetes.envoyDeployment.replicas`. Envoy Gateway stops reconciling that field when the EnvoyProxy leaves it out, which makes a manual scale to zero permanent and the Gateway unreachable for good. A `spec` that names its own replicas wins over this, and `null` leaves the field out entirely. Ignored for a spec that scales some other way - an `envoyHpa`, an `envoyDaemonSet`, or a `Host` provider. |
 | envoyProxy.spec | object | `{}` |  |
 | gateway | object | `{"addresses":[],"allowedRoutes":{"namespaces":{"from":"All"}},"annotations":{},"className":"envoy","create":true,"labels":{},"listeners":[],"name":"theia-shared-gateway","namespace":"eduide-system","redirects":[]}` | ------------------------------------------------------------------------ |
 | gateway.create | bool | `true` | Create the shared Gateway. Set false only if you terminate and route traffic some other way; EduIDE then has no ingress of its own. |

charts/eduide-cluster/templates/gateway/envoyproxy.yaml

@@ -1,4 +1,26 @@
 {{- if .Values.envoyProxy.create }}
+{{- /*
+Envoy Gateway only owns the data plane Deployment's replica count while the
+EnvoyProxy states it. Omit it and the controller writes the field once at
+creation and never looks at it again, so a single `kubectl scale --replicas=0`
+- or a stray click in a cluster UI - takes every Gateway on the class down
+until somebody scales it back by hand. Default it here, and let a spec that
+sets its own replicas (an HPA-driven cluster, say) win.
+
+Three specs must be left alone. A DaemonSet data plane rejects the resource
+outright - the CRD permits envoyDeployment or envoyDaemonSet, never both - an
+HPA already owns the field, and a `Host` provider has no Deployment to scale.
+`type` comes along with the default because the CRD requires it as soon as
+`provider` exists at all, and a spec of `{}` is a valid install.
+*/}}
+{{- $spec := deepCopy (default (dict) .Values.envoyProxy.spec) }}
+{{- $kubernetes := default (dict) (dig "provider" "kubernetes" (dict) $spec) }}
+{{- $onKubernetes := eq (dig "provider" "type" "Kubernetes" $spec) "Kubernetes" }}
+{{- $ownsReplicas := or (hasKey $kubernetes "envoyDaemonSet") (hasKey $kubernetes "envoyHpa") }}
+{{- if and (not (kindIs "invalid" .Values.envoyProxy.replicas)) $onKubernetes (not $ownsReplicas) }}
+{{- $default := dict "provider" (dict "type" "Kubernetes" "kubernetes" (dict "envoyDeployment" (dict "replicas" .Values.envoyProxy.replicas))) }}
+{{- $spec = mergeOverwrite $default $spec }}
+{{- end }}
 apiVersion: gateway.envoyproxy.io/v1alpha1
 kind: EnvoyProxy
 metadata:
@@ -13,5 +35,5 @@ metadata:
 {{ toYaml . | nindent 4 }}
   {{- end }}
 spec:
-{{ toYaml .Values.envoyProxy.spec | nindent 2 }}
+{{ toYaml $spec | nindent 2 }}
 {{- end }}

charts/eduide-cluster/values.yaml

@@ -133,6 +133,14 @@ envoyProxy:
   namespace: envoy-gateway-system
   labels: {}
   annotations: {}
+  # -- Data plane replica count, rendered into the spec at
+  # `provider.kubernetes.envoyDeployment.replicas`. Envoy Gateway stops
+  # reconciling that field when the EnvoyProxy leaves it out, which makes a
+  # manual scale to zero permanent and the Gateway unreachable for good. A
+  # `spec` that names its own replicas wins over this, and `null` leaves the
+  # field out entirely. Ignored for a spec that scales some other way - an
+  # `envoyHpa`, an `envoyDaemonSet`, or a `Host` provider.
+  replicas: 1
   spec: {}
 managedCertificates:
   enabled: false