fix(helm): preserve STS serviceName + networkPolicy.egress back-compat

helm/sim/.claude/skills/sim-helm/references/values-model.md

@@ -44,11 +44,9 @@ The Sim chart splits configuration across **four** layers. Understanding which l
 
 ## Why this layering exists
 
-**Single source of truth per concern.** Secrets live in a Secret. Operational defaults live where users can override them. Chart-computed values live where the chart can authoritatively compute them.
+**ESO compatibility.** When `externalSecrets.enabled=true`, the chart-managed Secret is **not rendered** — ESO renders one instead. Anything in Layer 1 must be mapped via `remoteRefs.app.<KEY>` or it's silently missing. Layers 2–4 are unaffected by ESO.
 
-**ESO compatibility.** When `externalSecrets.enabled=true`, the chart-managed Secret is **not rendered** — ESO renders one instead. Anything in Layer 1 must be mapped via `remoteRefs.app.<KEY>` or it's silently missing. Layers 2–4 are unaffected by ESO. Putting operational tunables in `envDefaults` instead of `env` means ESO users don't have to map dozens of tunables — just the real secrets.
-
-**Backwards compatibility.** Layer 2 was added in chart 1.0.0 (formerly all defaults lived in `app.env`). The override-skip logic in the Deployment template means existing users who set values in `app.env` continue to work — those values win over `envDefaults`.
+**Override precedence.** Values set in `app.env` (Layer 1 overrides) win over `envDefaults` (Layer 2) — so users who already had operational tunables in `app.env` continue to work.
 
 ## Where keys live — the canonical list
 

helm/sim/README.md

@@ -217,7 +217,7 @@ Before installing in production, confirm each of the following:
 * **Pinned images** — override `image.tag` (or `image.digest`) with an explicit version. Do not rely on the chart's default tag in production.
 * **Secrets management** — provide secrets via External Secrets Operator (ESO) or pre-created Kubernetes Secrets. Never commit secrets to `values.yaml`.
 * **TLS / Ingress** — set the `cert-manager.io/cluster-issuer` annotation on the ingress and tune `proxy-body-size` / `proxy-read-timeout` for your workload. See commented examples in `values.yaml`.
-* **Network policy egress** — review `networkPolicy.egress.exceptCidrs`. Defaults block cloud metadata endpoints (`169.254.169.254/32`, `169.254.170.2/32`); add your cluster's API server CIDR for stronger isolation.
+* **Network policy egress** — review `networkPolicy.egressExceptCidrs`. Defaults block cloud metadata endpoints (`169.254.169.254/32`, `169.254.170.2/32`); add your cluster's API server CIDR for stronger isolation. Custom egress rules go in `networkPolicy.egress` (a list).
 * **Namespace hardening** — label the install namespace with Pod Security Standards `restricted` enforcement (`pod-security.kubernetes.io/enforce=restricted`).
 * **Env validation** — keys under `app.env`, `realtime.env`, and `copilot.env` are passed through to the application and validated at startup. The JSON Schema intentionally does not enforce `additionalProperties: false` (would break custom user envs), so typos like `OPENA_API_KEY` (instead of `OPENAI_API_KEY`) surface as missing-key errors at runtime, not at `helm install` time. Review your env block carefully.
 * **Set public URLs** — `app.env.NEXT_PUBLIC_APP_URL` and `app.env.BETTER_AUTH_URL` must match your public origin (e.g. `https://sim.example.com`). Leaving them as `localhost` breaks sign-in.

helm/sim/templates/NOTES.txt

@@ -81,7 +81,16 @@ Your release is named {{ .Release.Name }} in namespace {{ .Release.Namespace }}.
      # Upgrade after changing values
      helm upgrade {{ .Release.Name }} ./helm/sim --namespace {{ .Release.Namespace }} -f your-values.yaml
 
-5. Where to go next:
+5. Upgrade notes (read before upgrading from a chart version released before this one):
+
+   * externalSecrets.apiVersion default is "v1beta1" (was "v1"). v1beta1 is
+     supported by every ESO release from v0.7+ through current. If you're on
+     ESO v0.17+ and want the graduated v1 API, set externalSecrets.apiVersion: "v1".
+   * networkPolicy.egress remains a list of custom egress rules (unchanged).
+     Cloud-metadata CIDR blocking is now configured via networkPolicy.egressExceptCidrs
+     (defaults to AWS/GCP/Azure IMDS + ECS task metadata).
+
+6. Where to go next:
 
    * Production checklist:  helm/sim/README.md (search "Production checklist")
    * Troubleshooting:       helm/sim/README.md (search "Troubleshooting")

helm/sim/templates/networkpolicy.yaml

@@ -107,14 +107,14 @@ spec:
     - ipBlock:
         cidr: 0.0.0.0/0
         except:
-          {{- range (default (list "169.254.169.254/32" "169.254.170.2/32") .Values.networkPolicy.egress.exceptCidrs) }}
+          {{- range (default (list "169.254.169.254/32" "169.254.170.2/32") .Values.networkPolicy.egressExceptCidrs) }}
           - {{ . | quote }}
           {{- end }}
     ports:
     - protocol: TCP
       port: 443
   # Allow custom egress rules
-  {{- with .Values.networkPolicy.egress.extraRules }}
+  {{- with .Values.networkPolicy.egress }}
   {{- toYaml . | nindent 2 }}
   {{- end }}
 
@@ -189,14 +189,14 @@ spec:
     - ipBlock:
         cidr: 0.0.0.0/0
         except:
-          {{- range (default (list "169.254.169.254/32" "169.254.170.2/32") .Values.networkPolicy.egress.exceptCidrs) }}
+          {{- range (default (list "169.254.169.254/32" "169.254.170.2/32") .Values.networkPolicy.egressExceptCidrs) }}
           - {{ . | quote }}
           {{- end }}
     ports:
     - protocol: TCP
       port: 443
   # Allow custom egress rules
-  {{- with .Values.networkPolicy.egress.extraRules }}
+  {{- with .Values.networkPolicy.egress }}
   {{- toYaml . | nindent 2 }}
   {{- end }}
 {{- end }}
@@ -296,11 +296,96 @@ spec:
     - ipBlock:
         cidr: 0.0.0.0/0
         except:
-          {{- range (default (list "169.254.169.254/32" "169.254.170.2/32") .Values.networkPolicy.egress.exceptCidrs) }}
+          {{- range (default (list "169.254.169.254/32" "169.254.170.2/32") .Values.networkPolicy.egressExceptCidrs) }}
           - {{ . | quote }}
           {{- end }}
     ports:
     - protocol: TCP
       port: 443
 {{- end }}
+
+{{- if .Values.telemetry.enabled }}
+---
+# Network Policy for OpenTelemetry Collector
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: {{ include "sim.fullname" . }}-otel-collector
+  namespace: {{ .Release.Namespace }}
+  labels:
+    {{- include "sim.labels" . | nindent 4 }}
+    app.kubernetes.io/component: telemetry
+spec:
+  podSelector:
+    matchLabels:
+      {{- include "sim.selectorLabels" . | nindent 6 }}
+      app.kubernetes.io/component: telemetry
+  policyTypes:
+  - Ingress
+  - Egress
+  ingress:
+  # OTLP from app
+  - from:
+    - podSelector:
+        matchLabels:
+          {{- include "sim.app.selectorLabels" . | nindent 10 }}
+    ports:
+    - protocol: TCP
+      port: 4317
+    - protocol: TCP
+      port: 4318
+  # OTLP from realtime
+  {{- if .Values.realtime.enabled }}
+  - from:
+    - podSelector:
+        matchLabels:
+          {{- include "sim.realtime.selectorLabels" . | nindent 10 }}
+    ports:
+    - protocol: TCP
+      port: 4317
+    - protocol: TCP
+      port: 4318
+  {{- end }}
+  # OTLP from copilot
+  {{- if .Values.copilot.enabled }}
+  - from:
+    - podSelector:
+        matchLabels:
+          {{- include "sim.selectorLabels" . | nindent 10 }}
+          app.kubernetes.io/component: copilot
+    ports:
+    - protocol: TCP
+      port: 4317
+    - protocol: TCP
+      port: 4318
+  {{- end }}
+  egress:
+  # DNS
+  - to: []
+    ports:
+    - protocol: UDP
+      port: 53
+    - protocol: TCP
+      port: 53
+  # HTTPS for forwarding to external observability backends (Datadog, Honeycomb, etc.)
+  - to:
+    - ipBlock:
+        cidr: 0.0.0.0/0
+        except:
+          {{- range (default (list "169.254.169.254/32" "169.254.170.2/32") .Values.networkPolicy.egressExceptCidrs) }}
+          - {{ . | quote }}
+          {{- end }}
+    ports:
+    - protocol: TCP
+      port: 443
+{{- end }}
+
+{{- /*
+  Copilot + copilot-postgresql intentionally do NOT ship dedicated NetworkPolicies.
+  Copilot requires REDIS_URL (external Redis on a non-443 port), and the chart
+  cannot know the user's Redis host/port at render time — a default egress rule
+  would silently block Redis on most installs. Users running networkPolicy.enabled=true
+  with copilot enabled should add their own NPs (or extend networkPolicy.egress
+  with the appropriate egress rules).
+*/}}
 {{- end }}

helm/sim/templates/statefulset-copilot-postgres.yaml

@@ -66,7 +66,11 @@ metadata:
     {{- include "sim.labels" . | nindent 4 }}
     app.kubernetes.io/component: copilot-postgresql
 spec:
-  serviceName: {{ include "sim.fullname" . }}-copilot-postgresql-headless
+  # Must remain {{ include "sim.fullname" . }}-copilot-postgresql (not the
+  # -headless name) — spec.serviceName is immutable on a StatefulSet, and
+  # the prior chart shipped with this value. Same rationale as the main
+  # postgresql STS; see statefulset-postgresql.yaml for details.
+  serviceName: {{ include "sim.fullname" . }}-copilot-postgresql
   replicas: 1
   podManagementPolicy: OrderedReady
   updateStrategy:
@@ -111,6 +115,10 @@ spec:
           envFrom:
             - secretRef:
                 name: {{ include "sim.fullname" . }}-copilot-postgresql-secret
+          {{- if .Values.copilot.postgresql.startupProbe }}
+          startupProbe:
+            {{- toYaml .Values.copilot.postgresql.startupProbe | nindent 12 }}
+          {{- end }}
           {{- if .Values.copilot.postgresql.livenessProbe }}
           livenessProbe:
             {{- toYaml .Values.copilot.postgresql.livenessProbe | nindent 12 }}

helm/sim/templates/statefulset-postgresql.yaml

@@ -90,7 +90,12 @@ metadata:
   labels:
     {{- include "sim.postgresql.labels" . | nindent 4 }}
 spec:
-  serviceName: {{ include "sim.fullname" . }}-postgresql-headless
+  # Must remain {{ include "sim.fullname" . }}-postgresql (not the -headless
+  # name) — spec.serviceName is immutable on a StatefulSet, and the prior
+  # chart shipped with this value. Changing it would break `helm upgrade` for
+  # every existing install with `Forbidden: updates to statefulset spec ...`.
+  # The headless Service in services.yaml is added alongside, not as a swap.
+  serviceName: {{ include "sim.fullname" . }}-postgresql
   replicas: 1
   minReadySeconds: 10
   podManagementPolicy: OrderedReady
@@ -135,6 +140,10 @@ spec:
                 name: {{ include "sim.fullname" . }}-postgresql-env
             - secretRef:
                 name: {{ include "sim.postgresqlSecretName" . }}
+          {{- if .Values.postgresql.startupProbe }}
+          startupProbe:
+            {{- toYaml .Values.postgresql.startupProbe | nindent 12 }}
+          {{- end }}
           {{- if .Values.postgresql.livenessProbe }}
           livenessProbe:
             {{- toYaml .Values.postgresql.livenessProbe | nindent 12 }}

helm/sim/tests/networkpolicy_test.yaml

@@ -128,10 +128,23 @@ tests:
               - protocol: TCP
                 port: 3000
 
-  - it: egress.extraRules are appended to both app and realtime NetworkPolicies
+  - it: telemetry collector NetworkPolicy renders when telemetry.enabled=true
     set:
       <<: *defaults
-      networkPolicy.egress.extraRules:
+      telemetry.enabled: true
+    documentIndex: 3
+    asserts:
+      - equal:
+          path: kind
+          value: NetworkPolicy
+      - equal:
+          path: metadata.name
+          value: t-sim-otel-collector
+
+  - it: networkPolicy.egress (custom rules) are appended to the app NetworkPolicy
+    set:
+      <<: *defaults
+      networkPolicy.egress:
         - to: []
           ports:
             - protocol: TCP
@@ -145,3 +158,21 @@ tests:
             ports:
               - protocol: TCP
                 port: 5432
+
+  - it: networkPolicy.egress (custom rules) are appended to the realtime NetworkPolicy
+    set:
+      <<: *defaults
+      networkPolicy.egress:
+        - to: []
+          ports:
+            - protocol: TCP
+              port: 5432
+    documentIndex: 1
+    asserts:
+      - contains:
+          path: spec.egress
+          content:
+            to: []
+            ports:
+              - protocol: TCP
+                port: 5432

helm/sim/values.yaml

@@ -622,12 +622,22 @@ postgresql:
     targetPort: 5432
 
   # Health checks
+  # startupProbe shields liveness from slow first-boot scenarios (pgvector
+  # extension init, WAL replay after a crash on a large data dir). Gives
+  # postgres up to 150s (30 * 5s) to become ready before liveness takes over.
+  startupProbe:
+    exec:
+      command: ["pg_isready", "-U", "postgres", "-d", "sim"]
+    periodSeconds: 5
+    failureThreshold: 30
+    timeoutSeconds: 5
+
   livenessProbe:
     exec:
       command: ["pg_isready", "-U", "postgres", "-d", "sim"]
     initialDelaySeconds: 10
     periodSeconds: 5
-  
+
   readinessProbe:
     exec:
       command: ["pg_isready", "-U", "postgres", "-d", "sim"]
@@ -954,7 +964,7 @@ monitoring:
 # to each other and to required external services (DNS, HTTPS) while blocking
 # everything else. The egress block additionally blacklists cloud metadata
 # endpoints (169.254.169.254/32, 169.254.170.2/32) by default — extend
-# egress.exceptCidrs with your cluster's API server CIDR for tighter isolation.
+# egressExceptCidrs with your cluster's API server CIDR for tighter isolation.
 # Your CNI must support NetworkPolicy (Calico, Cilium, GKE Dataplane V2, etc.).
 networkPolicy:
   enabled: false
@@ -973,16 +983,18 @@ networkPolicy:
   # Custom ingress rules appended to the policy
   ingress: []
 
-  # Egress configuration
-  egress:
-    # CIDRs excluded from broad HTTPS (443) egress.
-    # Defaults block AWS/GCP/Azure IMDS (169.254.169.254/32) and ECS task metadata
-    # (169.254.170.2/32). Add your cluster's API server CIDR for stronger isolation.
-    exceptCidrs:
-      - "169.254.169.254/32"
-      - "169.254.170.2/32"
-    # Custom egress rules appended to the policy
-    extraRules: []
+  # Custom egress rules appended to the policy.
+  # Kept as a top-level list (not a map) for backward compatibility with the
+  # pre-1.0 chart that shipped `networkPolicy.egress: []`. Existing values
+  # files continue to work without changes.
+  egress: []
+
+  # CIDRs excluded from broad HTTPS (443) egress.
+  # Defaults block AWS/GCP/Azure IMDS (169.254.169.254/32) and ECS task metadata
+  # (169.254.170.2/32). Add your cluster's API server CIDR for stronger isolation.
+  egressExceptCidrs:
+    - "169.254.169.254/32"
+    - "169.254.170.2/32"
 
 # Shared storage for enterprise workflows requiring data sharing between pods
 sharedStorage:
@@ -1438,14 +1450,24 @@ copilot:
       targetPort: 5432
 
     # Health checks
+    # startupProbe shields liveness from slow first-boot scenarios (pgvector
+    # extension init, WAL replay after a crash). Gives postgres up to 150s
+    # (30 * 5s) to become ready before liveness takes over.
+    startupProbe:
+      exec:
+        command: ["pg_isready", "-U", "copilot", "-d", "copilot"]
+      periodSeconds: 5
+      failureThreshold: 30
+      timeoutSeconds: 5
+
     livenessProbe:
       exec:
         command: ["pg_isready", "-U", "copilot", "-d", "copilot"]
       initialDelaySeconds: 10
       periodSeconds: 5
       timeoutSeconds: 5
       failureThreshold: 10
-    
+
     readinessProbe:
       exec:
         command: ["pg_isready", "-U", "copilot", "-d", "copilot"]