Enable Local HA

docs/operator-public-documentation/preview/index.md

@@ -404,6 +404,38 @@ The DocumentDB operator uses a sidecar injector plugin to automatically inject t
 For detailed information on configuring the sidecar injector plugin, see: [Sidecar Injector Plugin Configuration](../../developer-guides/sidecar-injector-plugin-configuration.md)
 
 
+### Local High-Availability (HA)
+
+The DocumentDB operator supports local high-availability by deploying multiple DocumentDB instances with automatic failover capabilities. This is achieved by setting `instancesPerNode` to a value greater than 1, which creates a cluster of DocumentDB instances with built-in redundancy.
+
+#### Enable Local HA
+
+To enable local HA, set `instancesPerNode: 3` in your DocumentDB resource:
+
+```sh
+cat <<EOF | kubectl apply -f -
+apiVersion: db.microsoft.com/preview
+kind: DocumentDB
+metadata:
+  name: documentdb-ha
+  namespace: your-namespace
+spec:
+  nodeCount: 1
+  instancesPerNode: 3  # Creates 3 DocumentDB instances for HA
+  documentDbCredentialSecret: documentdb-credentials
+  resource:
+    storage:
+      pvcSize: 10Gi
+  exposeViaService:
+    serviceType: LoadBalancer
+EOF
+```
+
+This configuration creates:
+- **1 Primary instance**: Handles all write operations
+- **2 Replica instances**: Provide read scalability and automatic failover capability
+
+
 ### Multi-Cloud Deployment
 
 The DocumentDB operator supports deployment across multiple cloud environments and Kubernetes distributions. For guidance on multi-cloud deployments, see: [Multi-Cloud Deployment Guide](../../../documentdb-playground/multi-clould-setup/multi-cloud-deployment-guide.md)

operator/documentdb-helm-chart/crds/db.microsoft.com_documentdbs.yaml

@@ -176,9 +176,9 @@ spec:
                   If not specified, defaults to a version that matches the DocumentDB operator version.
                 type: string
               instancesPerNode:
-                description: InstancesPerNode is the number of DocumentDB instances
-                  per node. Must be 1.
-                maximum: 1
+                description: 'InstancesPerNode is the number of DocumentDB instances
+                  per node. Range: 1-3.'
+                maximum: 3
                 minimum: 1
                 type: integer
               logLevel:

operator/src/api/preview/documentdb_types.go

@@ -15,9 +15,9 @@ type DocumentDBSpec struct {
 	// +kubebuilder:validation:Maximum=1
 	NodeCount int `json:"nodeCount"`
 
-	// InstancesPerNode is the number of DocumentDB instances per node. Must be 1.
+	// InstancesPerNode is the number of DocumentDB instances per node. Range: 1-3.
 	// +kubebuilder:validation:Minimum=1
-	// +kubebuilder:validation:Maximum=1
+	// +kubebuilder:validation:Maximum=3
 	InstancesPerNode int `json:"instancesPerNode"`
 
 	// Resource specifies the storage resources for DocumentDB.

operator/src/config/crd/bases/db.microsoft.com_documentdbs.yaml

@@ -176,9 +176,9 @@ spec:
                   If not specified, defaults to a version that matches the DocumentDB operator version.
                 type: string
               instancesPerNode:
-                description: InstancesPerNode is the number of DocumentDB instances
-                  per node. Must be 1.
-                maximum: 1
+                description: 'InstancesPerNode is the number of DocumentDB instances
+                  per node. Range: 1-3.'
+                maximum: 3
                 minimum: 1
                 type: integer
               logLevel:

operator/src/internal/utils/util.go

@@ -36,9 +36,8 @@ func GetDocumentDBServiceDefinition(documentdb *dbpreview.DocumentDB, replicatio
 	}
 	if replicationContext.EndpointEnabled() {
 		selector = map[string]string{
-			LABEL_APP:          documentdb.Name,
-			LABEL_REPLICA_TYPE: "primary", // Service forwards traffic to primary replicas
-			LABEL_ROLE:         "primary",
+			"cnpg.io/cluster":      documentdb.Name,
+			"cnpg.io/instanceRole": "primary", // Service forwards traffic to CNPG primary instance
 		}
 	}
 

operator/src/internal/utils/util_test.go

@@ -5,6 +5,11 @@ package util
 
 import (
 	"testing"
+
+	dbpreview "github.com/microsoft/documentdb-operator/api/preview"
+	corev1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/types"
 )
 
 func TestGenerateServiceName(t *testing.T) {
@@ -82,3 +87,134 @@ func TestGenerateServiceName(t *testing.T) {
 		})
 	}
 }
+
+func TestGetDocumentDBServiceDefinition_CNPGLabels(t *testing.T) {
+	tests := []struct {
+		name             string
+		documentDBName   string
+		endpointEnabled  bool
+		serviceType      corev1.ServiceType
+		expectedSelector map[string]string
+		description      string
+	}{
+		{
+			name:            "endpoint disabled - should have disabled selector",
+			documentDBName:  "test-documentdb",
+			endpointEnabled: false,
+			serviceType:     corev1.ServiceTypeLoadBalancer,
+			expectedSelector: map[string]string{
+				"disabled": "true",
+			},
+			description: "When endpoint is disabled, service should have disabled selector",
+		},
+		{
+			name:            "endpoint enabled with LoadBalancer - should use CNPG labels",
+			documentDBName:  "test-documentdb",
+			endpointEnabled: true,
+			serviceType:     corev1.ServiceTypeLoadBalancer,
+			expectedSelector: map[string]string{
+				"cnpg.io/cluster":      "test-documentdb",
+				"cnpg.io/instanceRole": "primary",
+			},
+			description: "When endpoint is enabled, service should use CNPG labels for failover support",
+		},
+		{
+			name:            "endpoint enabled with ClusterIP - should use CNPG labels",
+			documentDBName:  "test-documentdb",
+			endpointEnabled: true,
+			serviceType:     corev1.ServiceTypeClusterIP,
+			expectedSelector: map[string]string{
+				"cnpg.io/cluster":      "test-documentdb",
+				"cnpg.io/instanceRole": "primary",
+			},
+			description: "Service type should not affect selector labels",
+		},
+		{
+			name:            "different documentdb name - should reflect in cluster label",
+			documentDBName:  "my-db-cluster",
+			endpointEnabled: true,
+			serviceType:     corev1.ServiceTypeLoadBalancer,
+			expectedSelector: map[string]string{
+				"cnpg.io/cluster":      "my-db-cluster",
+				"cnpg.io/instanceRole": "primary",
+			},
+			description: "Cluster label should match DocumentDB instance name",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			// Create a mock DocumentDB instance
+			documentdb := &dbpreview.DocumentDB{
+				TypeMeta: metav1.TypeMeta{
+					APIVersion: "db.microsoft.com/preview",
+					Kind:       "DocumentDB",
+				},
+				ObjectMeta: metav1.ObjectMeta{
+					Name:      tt.documentDBName,
+					Namespace: "test-namespace",
+					UID:       types.UID("test-uid-123"),
+				},
+			}
+
+			// Create a mock ReplicationContext
+			replicationContext := &ReplicationContext{
+				Self:        tt.documentDBName,
+				Environment: "test",
+				state:       NoReplication, // This will make EndpointEnabled() return true
+			}
+
+			// If endpoint should be disabled, set a different state
+			if !tt.endpointEnabled {
+				replicationContext.state = Primary
+				replicationContext.currentLocalPrimary = "different-primary"
+				replicationContext.targetLocalPrimary = "target-primary"
+			}
+
+			// Generate the service definition
+			service := GetDocumentDBServiceDefinition(documentdb, replicationContext, "test-namespace", tt.serviceType)
+
+			// Verify the selector matches expected values
+			if len(service.Spec.Selector) != len(tt.expectedSelector) {
+				t.Errorf("Expected selector to have %d labels, got %d. Expected: %v, Got: %v",
+					len(tt.expectedSelector), len(service.Spec.Selector), tt.expectedSelector, service.Spec.Selector)
+			}
+
+			for key, expectedValue := range tt.expectedSelector {
+				if actualValue, exists := service.Spec.Selector[key]; !exists {
+					t.Errorf("Expected selector to contain key %q, but it was missing. Selector: %v", key, service.Spec.Selector)
+				} else if actualValue != expectedValue {
+					t.Errorf("Expected selector[%q] = %q, got %q", key, expectedValue, actualValue)
+				}
+			}
+
+			// Verify other service properties
+			if service.Name == "" {
+				t.Error("Service name should not be empty")
+			}
+
+			if service.Namespace != "test-namespace" {
+				t.Errorf("Expected service namespace to be 'test-namespace', got %q", service.Namespace)
+			}
+
+			if service.Spec.Type != tt.serviceType {
+				t.Errorf("Expected service type to be %v, got %v", tt.serviceType, service.Spec.Type)
+			}
+
+			// Verify owner reference is set correctly
+			if len(service.OwnerReferences) != 1 {
+				t.Errorf("Expected 1 owner reference, got %d", len(service.OwnerReferences))
+			} else {
+				ownerRef := service.OwnerReferences[0]
+				if ownerRef.Name != tt.documentDBName {
+					t.Errorf("Expected owner reference name to be %q, got %q", tt.documentDBName, ownerRef.Name)
+				}
+				if ownerRef.Kind != "DocumentDB" {
+					t.Errorf("Expected owner reference kind to be 'DocumentDB', got %q", ownerRef.Kind)
+				}
+			}
+
+			t.Logf("✅ %s: %s", tt.name, tt.description)
+		})
+	}
+}