Merge pull request IBM#188 from IBM/helm-limits-check

Cleanup autoscaler rules and remove old unused chart

Signed-off-by: Mihai Criveti <crivetimihai@gmail.com>

Author: crivetimihai

.gitignore

@@ -1,3 +1,6 @@
+*.db
+*.bak
+*.backup
 mcpgateway.sbom.xml
 gateway_service_leader.lock
 docs/docs/test/

charts/README.md

@@ -168,13 +168,13 @@ helm lint .
 
 ```bash
 # Upgrade only the gateway image
-ahelm upgrade mcp-stack . -n mcp \
+ahelm upgrade mcp-stack . -n mcp-private\
   --set mcpContextForge.image.tag=v1.2.3 \
   --wait
 
 # Preview changes (requires helm‑diff plugin)
 helm plugin install https://github.com/databus23/helm-diff
-helm diff upgrade mcp-stack . -n mcp -f my-values.yaml
+helm diff upgrade mcp-stack . -n mcp-private-f my-values.yaml
 
 # Roll back to revision 1
 helm rollback mcp-stack 1 -n mcp
@@ -376,7 +376,7 @@ helm upgrade --install mcp-stack charts/mcp-stack \
 
 # Later: raise the ceiling & make scaling more aggressive
 helm upgrade mcp-stack charts/mcp-stack \
-  -n mcp \
+  -n mcp-private\
   --reuse-values \
   --set mcpContextForge.hpa.maxReplicas=20 \
   --set mcpContextForge.hpa.targetCPUUtilizationPercentage=60 \
@@ -391,12 +391,12 @@ Useful in emergencies or during load tests.
 
 ```bash
 # Bump minReplicas from 3 → 5
-kubectl patch hpa mcp-stack-mcpgateway -n mcp \
+kubectl patch hpa mcp-stack-mcpgateway -n mcp-private\
   --type merge \
   -p '{"spec":{"minReplicas":5}}'
 
 # Drop the CPU target from 80 % → 65 % (scale up sooner)
-kubectl patch hpa mcp-stack-mcpgateway -n mcp \
+kubectl patch hpa mcp-stack-mcpgateway -n mcp-private\
   --type json \
   -p '[{"op":"replace","path":"/spec/metrics/0/resource/target/averageUtilization","value":65}]'
 ```
@@ -421,6 +421,25 @@ NAME                   TARGETS          MINPODS   MAXPODS   REPLICAS
 mcp-stack-mcpgateway   55%/70%          2         15        4
 ```
 
+### Check scaling events
+
+```bash
+# 1. Show the last few scale-up / scale-down events
+kubectl describe hpa mcp-stack-mcpgateway -n mcp-private | tail -n 20
+
+# 2. Stream HPA events as they happen
+kubectl get events -n mcp-private \
+  --field-selector involvedObject.kind=HorizontalPodAutoscaler,\
+involvedObject.name=mcp-stack-mcpgateway \
+  --watch
+
+# 3. Watch target utilisation & replica count refresh every 2 s
+watch -n2 kubectl get hpa mcp-stack-mcpgateway -n mcp-private
+
+# 4. Live pod-level CPU / memory (confirm the numbers the HPA sees)
+kubectl top pods -l app=mcp-stack-mcpgateway -n mcp-private --sort-by=cpu
+```
+
 ---
 
 ### Prerequisites & Gotchas

charts/mcp-stack/values.yaml

@@ -12,12 +12,19 @@ global:
 ########################################################################
 mcpContextForge:
   replicaCount: 2                           # horizontal scaling for the gateway
-  hpa:                                      # Horizontal pod autoscaler
-    enabled: true                           # default on; switch to false to disable HPA
-    minReplicas: 2                          # must be >= 1
-    maxReplicas: 10
-    targetCPUUtilizationPercentage: 80      # average CPU utilisation target
-    targetMemoryUtilizationPercentage: 80   # uncomment to add memory metric
+
+  # --- HORIZONTAL POD AUTOSCALER --------------------------------------
+  # * Percentages compare live usage with the container *request* values
+  #   (limits are ignored by the HPA).
+  # * If both CPU and memory targets are set, crossing either threshold
+  #   triggers a scale event.
+  # --------------------------------------------------------------------
+  hpa:
+    enabled: true                           # Set to false to keep a fixed replica count
+    minReplicas: 2                          # Never scale below this
+    maxReplicas: 10                         # Never scale above this
+    targetCPUUtilizationPercentage: 90      # Scale up when avg CPU > 90 % of *request*
+    targetMemoryUtilizationPercentage: 90   # Scale up when avg memory > 90 % of *request*
 
   image:
     repository: ghcr.io/ibm/mcp-context-forge
@@ -68,7 +75,7 @@ mcpContextForge:
       memory: 1024Mi
     requests:
       cpu: 100m
-      memory: 256Mi
+      memory: 512Mi
 
   # Optional ingress for HTTP traffic
   ingress: