From 5bf2d9bcd69feb8a75bd47e666a956d189646d3a Mon Sep 17 00:00:00 2001
From: MarshHawk <9513841+MarshHawk@users.noreply.github.com>
Date: Wed, 3 Jun 2026 09:51:07 -0400
Subject: [PATCH] feat: support custom hpa metrics in litellm helm chart

---
 deploy/charts/litellm-helm/templates/hpa.yaml |   6 +
 .../charts/litellm-helm/tests/hpa_tests.yaml  | 207 ++++++++++++++++--
 deploy/charts/litellm-helm/values.yaml        |   1 +
 3 files changed, 192 insertions(+), 22 deletions(-)

diff --git a/deploy/charts/litellm-helm/templates/hpa.yaml b/deploy/charts/litellm-helm/templates/hpa.yaml
index fec4d1f5c5e..dc77ce8548e 100644
--- a/deploy/charts/litellm-helm/templates/hpa.yaml
+++ b/deploy/charts/litellm-helm/templates/hpa.yaml
@@ -16,7 +16,11 @@ spec:
   behavior:
     {{- toYaml .Values.autoscaling.behavior | nindent 4 }}
   {{- end }}
+  {{- if coalesce .Values.autoscaling.metrics .Values.autoscaling.targetCPUUtilizationPercentage .Values.autoscaling.targetMemoryUtilizationPercentage }}
   metrics:
+    {{- with .Values.autoscaling.metrics }}
+    {{- toYaml . | nindent 4 }}
+    {{- else }}
     {{- if .Values.autoscaling.targetCPUUtilizationPercentage }}
     - type: Resource
       resource:
@@ -33,4 +37,6 @@ spec:
           type: Utilization
           averageUtilization: {{ .Values.autoscaling.targetMemoryUtilizationPercentage }}
     {{- end }}
+    {{- end }}
+  {{- end }}
 {{- end }}
diff --git a/deploy/charts/litellm-helm/tests/hpa_tests.yaml b/deploy/charts/litellm-helm/tests/hpa_tests.yaml
index ec18c3591d3..e226782ddad 100644
--- a/deploy/charts/litellm-helm/tests/hpa_tests.yaml
+++ b/deploy/charts/litellm-helm/tests/hpa_tests.yaml
@@ -1,36 +1,199 @@
-suite: "hpa with behavior"
+suite: test hpa
 templates:
   - hpa.yaml
 tests:
-  - it: "renders behavior when set"
+  - it: should render custom autoscaling metrics and behavior
+    template: hpa.yaml
     set:
-      autoscaling.enabled: true
-      autoscaling.behavior:
-        scaleUp:
-          stabilizationWindowSeconds: 60
-          policies:
+      autoscaling:
+        enabled: true
+        minReplicas: 2
+        maxReplicas: 10
+        metrics:
+        - type: Object
+          object:
+            describedObject:
+              apiVersion: v1
+              kind: Service
+              name: litellm-helm
+            metric:
+              name: litellm_requests_per_second
+            target:
+              type: AverageValue
+              averageValue: "20"
+        behavior:
+          scaleUp:
+            stabilizationWindowSeconds: 0
+            selectPolicy: Max
+            policies:
+            - type: Percent
+              value: 100
+              periodSeconds: 60
+            - type: Pods
+              value: 4
+              periodSeconds: 60
+          scaleDown:
+            stabilizationWindowSeconds: 300
+            selectPolicy: Max
+            policies:
+            - type: Percent
+              value: 50
+              periodSeconds: 60
+    asserts:
+      - isKind:
+          of: HorizontalPodAutoscaler
+      - equal:
+          path: spec.minReplicas
+          value: 2
+      - equal:
+          path: spec.maxReplicas
+          value: 10
+      - equal:
+          path: spec.metrics
+          value:
+          - type: Object
+            object:
+              describedObject:
+                apiVersion: v1
+                kind: Service
+                name: litellm-helm
+              metric:
+                name: litellm_requests_per_second
+              target:
+                type: AverageValue
+                averageValue: "20"
+      - equal:
+          path: spec.behavior
+          value:
+            scaleUp:
+              stabilizationWindowSeconds: 0
+              selectPolicy: Max
+              policies:
+              - type: Percent
+                value: 100
+                periodSeconds: 60
+              - type: Pods
+                value: 4
+                periodSeconds: 60
+            scaleDown:
+              stabilizationWindowSeconds: 300
+              selectPolicy: Max
+              policies:
+              - type: Percent
+                value: 50
+                periodSeconds: 60
+
+  - it: should keep rendering cpu and memory metrics when custom metrics are not set
+    template: hpa.yaml
+    set:
+      autoscaling:
+        enabled: true
+        targetCPUUtilizationPercentage: 75
+        targetMemoryUtilizationPercentage: 70
+    asserts:
+      - isKind:
+          of: HorizontalPodAutoscaler
+      - equal:
+          path: spec.metrics
+          value:
+          - type: Resource
+            resource:
+              name: cpu
+              target:
+                type: Utilization
+                averageUtilization: 75
+          - type: Resource
+            resource:
+              name: memory
+              target:
+                type: Utilization
+                averageUtilization: 70
+      - isNull:
+          path: spec.behavior
+
+  - it: should prefer custom metrics over cpu and memory targets
+    template: hpa.yaml
+    set:
+      autoscaling:
+        enabled: true
+        targetCPUUtilizationPercentage: 75
+        targetMemoryUtilizationPercentage: 70
+        metrics:
+        - type: Object
+          object:
+            describedObject:
+              apiVersion: v1
+              kind: Service
+              name: litellm-helm
+            metric:
+              name: litellm_requests_per_second
+            target:
+              type: AverageValue
+              averageValue: "20"
+    asserts:
+      - isKind:
+          of: HorizontalPodAutoscaler
+      - equal:
+          path: spec.metrics
+          value:
+          - type: Object
+            object:
+              describedObject:
+                apiVersion: v1
+                kind: Service
+                name: litellm-helm
+              metric:
+                name: litellm_requests_per_second
+              target:
+                type: AverageValue
+                averageValue: "20"
+
+  - it: should render behavior when set
+    template: hpa.yaml
+    set:
+      autoscaling:
+        enabled: true
+        behavior:
+          scaleUp:
+            stabilizationWindowSeconds: 60
+            policies:
             - type: Pods
               value: 2
               periodSeconds: 60
-        scaleDown:
-          stabilizationWindowSeconds: 90
-          policies:
+          scaleDown:
+            stabilizationWindowSeconds: 90
+            policies:
             - type: Pods
               value: 1
               periodSeconds: 60
     asserts:
-      - isKind: { of: HorizontalPodAutoscaler }
-      - equal: { path: spec.behavior.scaleUp.stabilizationWindowSeconds, value: 60 }
-      - equal: { path: spec.behavior.scaleDown.stabilizationWindowSeconds, value: 90 }
+      - isKind:
+          of: HorizontalPodAutoscaler
+      - equal:
+          path: spec.behavior.scaleUp.stabilizationWindowSeconds
+          value: 60
+      - equal:
+          path: spec.behavior.scaleDown.stabilizationWindowSeconds
+          value: 90
 
----
-suite: "hpa without behavior"
-templates:
-  - hpa.yaml
-tests:
-  - it: "does not render behavior when not set"
+  - it: should omit metrics when no metric sources are set
+    template: hpa.yaml
     set:
-      autoscaling.enabled: true
+      autoscaling:
+        enabled: true
+        targetCPUUtilizationPercentage: null
+        targetMemoryUtilizationPercentage: null
+        metrics: []
     asserts:
-      - isKind: { of: HorizontalPodAutoscaler }
-      - isNull: { path: spec.behavior }
+      - isKind:
+          of: HorizontalPodAutoscaler
+      - isNull:
+          path: spec.metrics
+
+  - it: should not render when autoscaling is disabled
+    template: hpa.yaml
+    set:
+      autoscaling.enabled: false
+    asserts:
+      - hasDocuments:
+          count: 0
diff --git a/deploy/charts/litellm-helm/values.yaml b/deploy/charts/litellm-helm/values.yaml
index 81558ed5b29..e1f2a818926 100644
--- a/deploy/charts/litellm-helm/values.yaml
+++ b/deploy/charts/litellm-helm/values.yaml
@@ -184,6 +184,7 @@ autoscaling:
   maxReplicas: 100
   targetCPUUtilizationPercentage: 80
   # targetMemoryUtilizationPercentage: 80
+  metrics: []
   # behavior: {}
 
 # Autoscaling with keda is mutually exclusive with hpa