Merge pull request #42207 from BerriAI/litellm_helm_componentized_replica_count

fix(helm): render a fixed replicaCount on componentized deployments when HPA is disabled

helm/litellm/templates/backend/deployment.yaml

@@ -7,6 +7,9 @@ metadata:
     {{- include "litellm.commonLabels" . | nindent 4 }}
     app.kubernetes.io/component: backend
 spec:
+  {{- if and (not .Values.backend.hpa.enabled) (not (kindIs "invalid" .Values.backend.replicaCount)) }}
+  replicas: {{ .Values.backend.replicaCount }}
+  {{- end }}
   {{- with .Values.backend.strategy }}
   strategy:
     {{- toYaml . | nindent 4 }}

helm/litellm/templates/gateway/deployment.yaml

@@ -7,6 +7,9 @@ metadata:
     {{- include "litellm.commonLabels" . | nindent 4 }}
     app.kubernetes.io/component: gateway
 spec:
+  {{- if and (not .Values.gateway.hpa.enabled) (not (kindIs "invalid" .Values.gateway.replicaCount)) }}
+  replicas: {{ .Values.gateway.replicaCount }}
+  {{- end }}
   {{- with .Values.gateway.strategy }}
   strategy:
     {{- toYaml . | nindent 4 }}

helm/litellm/templates/ui/deployment.yaml

@@ -7,6 +7,9 @@ metadata:
     {{- include "litellm.commonLabels" . | nindent 4 }}
     app.kubernetes.io/component: ui
 spec:
+  {{- if and (not .Values.ui.hpa.enabled) (not (kindIs "invalid" .Values.ui.replicaCount)) }}
+  replicas: {{ .Values.ui.replicaCount }}
+  {{- end }}
   {{- with .Values.ui.strategy }}
   strategy:
     {{- toYaml . | nindent 4 }}

helm/litellm/tests/replica_count_tests.yaml

@@ -0,0 +1,100 @@
+suite: test fixed replica count when HPA is disabled
+templates:
+  - gateway/deployment.yaml
+  - gateway/configmap.yaml
+  - backend/deployment.yaml
+  - ui/deployment.yaml
+values:
+  - ./values/required.yaml
+tests:
+  - it: gateway renders replicaCount into spec.replicas when its HPA is disabled
+    template: gateway/deployment.yaml
+    set:
+      gateway.hpa.enabled: false
+      gateway.replicaCount: 3
+    asserts:
+      - isKind:
+          of: Deployment
+      - equal:
+          path: spec.replicas
+          value: 3
+
+  - it: backend renders replicaCount into spec.replicas when its HPA is disabled
+    template: backend/deployment.yaml
+    set:
+      backend.hpa.enabled: false
+      backend.replicaCount: 2
+    asserts:
+      - equal:
+          path: spec.replicas
+          value: 2
+
+  - it: ui renders replicaCount into spec.replicas when its HPA is disabled
+    template: ui/deployment.yaml
+    set:
+      ui.hpa.enabled: false
+      ui.replicaCount: 2
+    asserts:
+      - equal:
+          path: spec.replicas
+          value: 2
+
+  - it: replicaCount 0 scales the gateway to zero instead of being treated as unset
+    template: gateway/deployment.yaml
+    set:
+      gateway.hpa.enabled: false
+      gateway.replicaCount: 0
+    asserts:
+      - equal:
+          path: spec.replicas
+          value: 0
+
+  - it: a component with HPA disabled but no replicaCount set keeps omitting spec.replicas, so upgrades do not reset a hand-scaled Deployment
+    set:
+      gateway.hpa.enabled: false
+      backend.hpa.enabled: false
+      ui.hpa.enabled: false
+    asserts:
+      - notExists:
+          path: spec.replicas
+        template: gateway/deployment.yaml
+      - notExists:
+          path: spec.replicas
+        template: backend/deployment.yaml
+      - notExists:
+          path: spec.replicas
+        template: ui/deployment.yaml
+
+  - it: every component omits spec.replicas when its HPA is enabled, so the autoscaler owns the count
+    set:
+      gateway.hpa.enabled: true
+      gateway.replicaCount: 3
+      backend.hpa.enabled: true
+      backend.replicaCount: 3
+      ui.hpa.enabled: true
+      ui.replicaCount: 3
+    asserts:
+      - notExists:
+          path: spec.replicas
+        template: gateway/deployment.yaml
+      - notExists:
+          path: spec.replicas
+        template: backend/deployment.yaml
+      - notExists:
+          path: spec.replicas
+        template: ui/deployment.yaml
+
+  - it: a component with HPA disabled renders replicas while a sibling with HPA enabled does not
+    set:
+      gateway.hpa.enabled: false
+      gateway.replicaCount: 4
+      backend.hpa.enabled: true
+      backend.replicaCount: 4
+    asserts:
+      - equal:
+          path: spec.replicas
+          value: 4
+        template: gateway/deployment.yaml
+      - notExists:
+          path: spec.replicas
+        template: backend/deployment.yaml

helm/litellm/values.yaml

@@ -397,6 +397,11 @@ gateway:
   #   failureThreshold: 30
   #   periodSeconds: 10
   startupProbe: {}
+  # Optional fixed pod count, rendered into the Deployment's spec.replicas only
+  # when hpa.enabled is false. Unset by default so an existing Deployment keeps
+  # its current count; with the HPA on, the autoscaler owns the count, e.g.:
+  #   replicaCount: 3
+  replicaCount:
   hpa:
     enabled: true
     minReplicas: 1
@@ -524,6 +529,8 @@ backend:
   strategy: {}
   # Optional startupProbe; same shape as gateway.startupProbe. Empty by default.
   startupProbe: {}
+  # Same semantics as gateway.replicaCount.
+  replicaCount:
   hpa:
     enabled: true
     minReplicas: 1
@@ -590,6 +597,8 @@ ui:
   strategy: {}
   # Optional startupProbe; same shape as gateway.startupProbe. Empty by default.
   startupProbe: {}
+  # Same semantics as gateway.replicaCount.
+  replicaCount:
   hpa:
     enabled: false
     minReplicas: 1