SMOODEV-1965: smooai-next chart — optional hpa.memoryTarget for memory scale-out

helm/smooai-next/Chart.yaml

@@ -23,7 +23,11 @@ type: application
 #        (SMOODEV-1790 — dogfooding the chart for smooai apps/web).
 # 0.2.1: PDB sets unhealthyPodEvictionPolicy (default AlwaysAllow) — kube-linter
 #        flags its absence; matches the prior hand-rolled PDB.
-version: 0.2.1
+# 0.2.2: expose hpa.memoryTarget — optional second HPA Resource metric (memory
+#        averageUtilization) so memory-pressured Next.js pods scale out instead
+#        of riding into a V8 heap-limit OOM (SMOODEV-1965). Default 0 = CPU-only,
+#        backward-compatible.
+version: 0.2.2
 # Tracks the consuming app image version (override via image.tag).
 appVersion: "0.1.0"
 home: https://github.com/SmooAI/deploy

helm/smooai-next/README.md

@@ -28,7 +28,7 @@ helm/smooai-next/
 │   ├── _helpers.tpl
 │   ├── deployment.yaml               # node server.js; graceful shutdown; /api/health probes; S3 cache env
 │   ├── service.yaml                  # ClusterIP, port → http (3000)
-│   ├── hpa.yaml                      # HPA (min/max/cpuTarget)
+│   ├── hpa.yaml                      # HPA (min/max/cpuTarget; optional memoryTarget)
 │   ├── pdb.yaml                      # PodDisruptionBudget (minAvailable)
 │   ├── serviceaccount.yaml           # IRSA-annotated ServiceAccount
 │   ├── ingress.yaml                  # dedicated internet-facing ALB = the CloudFront origin

helm/smooai-next/templates/hpa.yaml

@@ -19,6 +19,19 @@ spec:
         target:
           type: Utilization
           averageUtilization: {{ .Values.hpa.cpuTarget }}
+    {{- with .Values.hpa.memoryTarget }}
+    # Optional memory-based scale-out (averageUtilization is % of the memory
+    # REQUEST). The HPA scales UP when EITHER metric is hot and DOWN only when
+    # BOTH are cool — so memory-pressured Next.js pods add replicas instead of
+    # riding individual pods into a V8 heap-limit OOM. `with` skips this when
+    # memoryTarget is unset/0, keeping the default CPU-only behavior unchanged.
+    - type: Resource
+      resource:
+        name: memory
+        target:
+          type: Utilization
+          averageUtilization: {{ . }}
+    {{- end }}
   {{- with .Values.hpa.behavior }}
   # Optional v2 scaling behavior (scaleUp/scaleDown stabilization + policies).
   behavior:

helm/smooai-next/values.yaml

@@ -45,6 +45,14 @@ hpa:
   min: 2
   max: 10
   cpuTarget: 70
+  # Optional memory-based scale-out target — averageUtilization as a % of the
+  # memory REQUEST (resources.requests.memory below). Unset/0 = CPU-only (the
+  # default; backward-compatible). Set e.g. 80 to add a second HPA metric so a
+  # memory-pressured Next.js workload scales out instead of riding each pod into
+  # a V8 heap-limit OOM. The HPA scales UP when EITHER cpu or memory is hot and
+  # DOWN only when BOTH are cool, so set this comfortably ABOVE steady-state
+  # (memory-used ÷ memory-request) or it will pin the deployment at max.
+  memoryTarget: 0
   # Optional autoscaling/v2 scaling behavior (scaleUp/scaleDown stabilization
   # windows + policies). Empty = the autoscaler's built-in defaults. Example:
   #   behavior: