foo

Author: pantierra

charts/eoapi/templates/knative-init.yaml

@@ -32,9 +32,80 @@ spec:
           set -e
           echo "=== Knative Initialization ==="
 
-          # Wait for knative-operator to be ready
-          echo "Waiting for knative-operator..."
-          kubectl rollout status deployment/knative-operator -n default --timeout={{ .Values.knative.initTimeout | default "600s" }}
+          # Wait for knative-operator with exponential backoff
+          echo "Waiting for knative-operator deployment to be available..."
+          RELEASE_NAMESPACE="{{ .Release.Namespace }}"
+          OPERATOR_NAMESPACE=""
+          max_attempts=60
+          attempt=1
+          backoff=5
+
+          while [ $attempt -le $max_attempts ]; do
+            echo "Attempt $attempt/$max_attempts: Looking for knative-operator..."
+
+            # First check the release namespace (most likely location for Helm dependency)
+            if kubectl get deployment knative-operator -n "$RELEASE_NAMESPACE" >/dev/null 2>&1; then
+              OPERATOR_NAMESPACE="$RELEASE_NAMESPACE"
+              echo "✅ Found knative-operator in release namespace: $OPERATOR_NAMESPACE"
+              break
+            fi
+
+            # Fallback: check all namespaces
+            OPERATOR_NAMESPACE=$(kubectl get deployment knative-operator --all-namespaces -o jsonpath='{.items[0].metadata.namespace}' 2>/dev/null || echo "")
+            if [ -n "$OPERATOR_NAMESPACE" ]; then
+              echo "✅ Found knative-operator in namespace: $OPERATOR_NAMESPACE"
+              break
+            fi
+
+            echo "⏳ knative-operator not found, waiting ${backoff}s before retry..."
+            sleep $backoff
+            attempt=$((attempt + 1))
+
+            # Exponential backoff with max of 30s
+            if [ $backoff -lt 30 ]; then
+              backoff=$((backoff * 2))
+            fi
+          done
+
+          if [ -z "$OPERATOR_NAMESPACE" ]; then
+            echo "❌ knative-operator deployment not found after $max_attempts attempts"
+            echo ""
+            echo "=== Debugging Information ==="
+            echo "Current namespace: $RELEASE_NAMESPACE"
+            echo ""
+            echo "Helm releases in current namespace:"
+            helm list -n "$RELEASE_NAMESPACE" || echo "No helm releases found"
+            echo ""
+            echo "All deployments in current namespace:"
+            kubectl get deployments -n "$RELEASE_NAMESPACE" -o wide || echo "No deployments found"
+            echo ""
+            echo "All deployments across all namespaces (first 20):"
+            kubectl get deployments --all-namespaces | head -21
+            echo ""
+            echo "Knative-related resources:"
+            kubectl get all --all-namespaces | grep -i knative || echo "No knative resources found"
+            echo ""
+            echo "Recent events in current namespace:"
+            kubectl get events -n "$RELEASE_NAMESPACE" --sort-by='.lastTimestamp' | tail -10 || echo "No events found"
+            echo ""
+            echo "Pod status in current namespace:"
+            kubectl get pods -n "$RELEASE_NAMESPACE" -o wide || echo "No pods found"
+            exit 1
+          fi
+
+          echo "Waiting for knative-operator deployment to be ready..."
+          if ! kubectl rollout status deployment/knative-operator -n "$OPERATOR_NAMESPACE" --timeout=600s; then
+            echo "❌ knative-operator failed to become ready within timeout"
+            echo "Deployment status:"
+            kubectl describe deployment knative-operator -n "$OPERATOR_NAMESPACE"
+            echo "Pod status:"
+            kubectl get pods -n "$OPERATOR_NAMESPACE" -l name=knative-operator -o wide
+            echo "Recent events:"
+            kubectl get events -n "$OPERATOR_NAMESPACE" --sort-by='.lastTimestamp' | tail -10
+            exit 1
+          fi
+
+          echo "✅ knative-operator is ready, proceeding with Knative setup..."
 
           # Create namespaces
           kubectl create namespace knative-serving --dry-run=client -o yaml | kubectl apply -f -
@@ -55,7 +126,8 @@ spec:
             echo "✅ KnativeServing created successfully"
           else
             echo "❌ Failed to create KnativeServing. Checking operator status..."
-            kubectl get pods -n default -l name=knative-operator
+            kubectl get pods -n "$OPERATOR_NAMESPACE" -l name=knative-operator
+            kubectl logs -n "$OPERATOR_NAMESPACE" -l name=knative-operator --tail=20
             exit 1
           fi
 
@@ -74,7 +146,8 @@ spec:
             echo "✅ KnativeEventing created successfully"
           else
             echo "❌ Failed to create KnativeEventing. Checking operator status..."
-            kubectl get pods -n default -l name=knative-operator
+            kubectl get pods -n "$OPERATOR_NAMESPACE" -l name=knative-operator
+            kubectl logs -n "$OPERATOR_NAMESPACE" -l name=knative-operator --tail=20
             exit 1
           fi
 
@@ -87,6 +160,6 @@ spec:
           limits:
             cpu: 50m
             memory: 64Mi
-  backoffLimit: 3
-  activeDeadlineSeconds: 600
+  backoffLimit: 1
+  activeDeadlineSeconds: 1200
 {{- end }}