Add ray head service endpoints in status (#341)

Co-authored-by: chenyu.jiang <[email protected]>

apiserver/pkg/model/converter.go

@@ -51,6 +51,10 @@ func FromCrdToApiCluster(cluster *v1alpha1.RayCluster, events []v1.Event) *api.C
 		pbCluster.Events = append(pbCluster.Events, clusterEvent)
 	}
 
+	pbCluster.ServiceEndpoint = map[string]string{}
+	for name, port := range cluster.Status.Endpoints {
+		pbCluster.ServiceEndpoint[name] = port
+	}
 	return pbCluster
 }
 

proto/cluster.proto

@@ -158,6 +158,9 @@ message Cluster {
 
   // Output. The list related to the cluster.
   repeated ClusterEvent events = 10;
+
+  // Output. The service endpoint of the cluster
+  map<string, string> service_endpoint = 11;
 }
 
 message ClusterSpec {

proto/swagger/cluster.swagger.json

@@ -266,6 +266,13 @@
             "$ref": "#/definitions/protoClusterEvent"
           },
           "description": "Output. The list related to the cluster."
+        },
+        "serviceEndpoint": {
+          "type": "object",
+          "additionalProperties": {
+            "type": "string"
+          },
+          "title": "Output. The service endpoint of the cluster"
         }
       }
     },

ray-operator/apis/ray/v1alpha1/raycluster_types.go

@@ -111,6 +111,8 @@ type RayClusterStatus struct {
 	// LastUpdateTime indicates last update timestamp for this cluster status.
 	// +nullable
 	LastUpdateTime *metav1.Time `json:"lastUpdateTime,omitempty"`
+	// Service Endpoints
+	Endpoints map[string]string `json:"endpoints,omitempty"`
 }
 
 // RayNodeType  the type of a ray node: head/worker

ray-operator/config/crd/bases/ray.io_rayclusters.yaml

@@ -10987,6 +10987,11 @@ spec:
                   claimed by the user at the cluster level.
                 format: int32
                 type: integer
+              endpoints:
+                additionalProperties:
+                  type: string
+                description: Service Endpoints
+                type: object
               lastUpdateTime:
                 description: LastUpdateTime indicates last update timestamp for this
                   cluster status.

ray-operator/config/crd/bases/ray.io_rayservices.yaml

@@ -11589,6 +11589,11 @@ spec:
                           replicas claimed by the user at the cluster level.
                         format: int32
                         type: integer
+                      endpoints:
+                        additionalProperties:
+                          type: string
+                        description: Service Endpoints
+                        type: object
                       lastUpdateTime:
                         description: LastUpdateTime indicates last update timestamp
                           for this cluster status.
@@ -11683,6 +11688,11 @@ spec:
                           replicas claimed by the user at the cluster level.
                         format: int32
                         type: integer
+                      endpoints:
+                        additionalProperties:
+                          type: string
+                        description: Service Endpoints
+                        type: object
                       lastUpdateTime:
                         description: LastUpdateTime indicates last update timestamp
                           for this cluster status.

ray-operator/controllers/ray/raycluster_controller.go

@@ -634,6 +634,30 @@ func (r *RayClusterReconciler) updateStatus(instance *rayiov1alpha1.RayCluster)
 		}
 	}
 
+	rayHeadSvc := corev1.ServiceList{}
+	filterLabels = client.MatchingLabels{
+		common.RayClusterLabelKey:  instance.Name,
+		common.RayNodeTypeLabelKey: "head",
+	}
+	// TODO: (@scarlet25151) for now there would be several kubernetes serivces related to
+	// one raycluster, we have used the label to select the headservice and pick the first one.
+	// we may need use Get method to select by name.
+	if err := r.List(context.TODO(), &rayHeadSvc, client.InNamespace(instance.Namespace), filterLabels); err != nil {
+		return err
+	}
+	if len(rayHeadSvc.Items) != 0 {
+		svc := rayHeadSvc.Items[0]
+		if instance.Status.Endpoints == nil {
+			instance.Status.Endpoints = map[string]string{}
+		}
+		for _, port := range svc.Spec.Ports {
+			if len(port.Name) == 0 {
+				r.Log.Info("updateStatus", "service port name is empty", port)
+				continue
+			}
+			instance.Status.Endpoints[port.Name] = fmt.Sprintf("%d", port.NodePort)
+		}
+	}
 	timeNow := metav1.Now()
 	instance.Status.LastUpdateTime = &timeNow
 	if err := r.Status().Update(context.Background(), instance); err != nil {