[apiserver] Start apiserver v2 in apiserver/cmd/main.go

apiserver/Dockerfile

@@ -14,6 +14,7 @@ RUN go mod download
 COPY proto/ proto/
 COPY ray-operator/ ray-operator/
 COPY apiserver/ apiserver/
+COPY apiserversdk/ apiserversdk/
 
 WORKDIR /workspace/apiserver
 

apiserver/cmd/main.go

@@ -23,12 +23,14 @@ import (
 	"google.golang.org/grpc/reflection"
 	"google.golang.org/protobuf/encoding/protojson"
 	"k8s.io/klog/v2"
+	"sigs.k8s.io/controller-runtime/pkg/client/config"
 
 	"github.com/ray-project/kuberay/apiserver/pkg/interceptor"
 	"github.com/ray-project/kuberay/apiserver/pkg/manager"
 	"github.com/ray-project/kuberay/apiserver/pkg/server"
 	"github.com/ray-project/kuberay/apiserver/pkg/swagger"
 	"github.com/ray-project/kuberay/apiserver/pkg/util"
+	"github.com/ray-project/kuberay/apiserversdk"
 	api "github.com/ray-project/kuberay/proto/go_client"
 )
 
@@ -39,6 +41,7 @@ var (
 	logFile            = flag.String("logFilePath", "", "Synchronize logs to local file")
 	localSwaggerPath   = flag.String("localSwaggerPath", "", "Specify the root directory for `*.swagger.json` the swagger files.")
 	grpcTimeout        = flag.Duration("grpc_timeout", util.GRPCServerDefaultTimeout, "gRPC server timeout duration")
+	enableAPIServerV2  = flag.Bool("enable-api-server-v2", true, "Enable API server V2")
 	healthy            int32
 )
 
@@ -145,7 +148,23 @@ func startHttpProxy() {
 	registerHttpHandlerFromEndpoint(ctx, api.RegisterRayJobSubmissionServiceHandlerFromEndpoint, "RayJobSubmissionService", runtimeMux)
 
 	// Create a top level mux to include both Http gRPC servers and other endpoints like metrics
-	topMux := http.NewServeMux()
+	var topMux *http.ServeMux
+	if *enableAPIServerV2 {
+		kubernetesConfig, err := config.GetConfig()
+		if err != nil {
+			klog.Fatalf("Failed to load kubeconfig: %v", err)
+		}
+
+		topMux, err = apiserversdk.NewMux(apiserversdk.MuxConfig{
+			KubernetesConfig: kubernetesConfig,
+		})
+		if err != nil {
+			klog.Fatalf("Failed to create API server mux: %v", err)
+		}
+	} else {
+		topMux = http.NewServeMux()
+	}
+
 	// Seems /apis (matches /apis/v1alpha1/clusters) works fine
 	topMux.Handle("/", runtimeMux)
 	topMux.Handle("/metrics", promhttp.Handler())

apiserver/deploy/base/insecure/apiserver.yaml

@@ -133,6 +133,32 @@ rules:
   - get
   - list
 ---
+# apiserversdk requires the following permissions to be able to list and proxy services
+apiVersion: rbac.authorization.k8s.io/v1
+kind: ClusterRole
+metadata:
+  name: allow-service-access
+rules:
+  - apiGroups: [""]
+    resources: ["services"]
+    verbs: ["get", "watch", "list"]
+  - apiGroups: [""]
+    resources: ["services/proxy"]
+    verbs: ["get", "watch", "list"]
+---
+apiVersion: rbac.authorization.k8s.io/v1
+kind: ClusterRoleBinding
+metadata:
+  name: allow-service-access
+subjects:
+  - kind: ServiceAccount
+    name: kuberay-apiserver
+    namespace: ray-system
+roleRef:
+  kind: ClusterRole
+  name: allow-service-access
+  apiGroup: rbac.authorization.k8s.io
+---
 apiVersion: v1
 kind: Namespace
 metadata:

apiserver/deploy/base/secure/apiserver.yaml

@@ -164,6 +164,32 @@ rules:
   - get
   - list
 ---
+# apiserversdk requires the following permissions to be able to list and proxy services
+apiVersion: rbac.authorization.k8s.io/v1
+kind: ClusterRole
+metadata:
+  name: allow-service-access
+rules:
+  - apiGroups: [""]
+    resources: ["services"]
+    verbs: ["get", "watch", "list"]
+  - apiGroups: [""]
+    resources: ["services/proxy"]
+    verbs: ["get", "watch", "list"]
+---
+apiVersion: rbac.authorization.k8s.io/v1
+kind: ClusterRoleBinding
+metadata:
+  name: allow-service-access
+subjects:
+  - kind: ServiceAccount
+    name: kuberay-apiserver
+    namespace: ray-system
+roleRef:
+  kind: ClusterRole
+  name: allow-service-access
+  apiGroup: rbac.authorization.k8s.io
+---
 apiVersion: v1
 kind: Namespace
 metadata:

apiserver/test/e2e/apiserversdk/cluster_e2e_test.go

@@ -0,0 +1,45 @@
+package apiserversdk
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/require"
+	corev1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+
+	rayv1 "github.com/ray-project/kuberay/ray-operator/apis/ray/v1"
+)
+
+func TestCreateCluster(t *testing.T) {
+	tCtx, err := NewEnd2EndTestingContext(t)
+	require.NoError(t, err, "No error expected when creating testing context")
+	rayClient := tCtx.GetRayHttpClient()
+	rayCluster := &rayv1.RayCluster{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      tCtx.GetRayClusterName(),
+			Namespace: tCtx.GetNamespaceName(),
+		},
+		Spec: rayv1.RayClusterSpec{
+			HeadGroupSpec: rayv1.HeadGroupSpec{
+				RayStartParams: map[string]string{},
+				Template: corev1.PodTemplateSpec{
+					Spec: corev1.PodSpec{
+						Containers: []corev1.Container{
+							{Name: "test", Image: "test"},
+						},
+					},
+				},
+			},
+		},
+	}
+	_, err = rayClient.RayClusters(tCtx.GetNamespaceName()).Create(tCtx.GetCtx(), rayCluster, metav1.CreateOptions{})
+	require.NoError(t, err)
+
+	actualCluster, err := rayClient.RayClusters(tCtx.GetNamespaceName()).Get(tCtx.GetCtx(), tCtx.GetRayClusterName(), metav1.GetOptions{})
+	require.NoError(t, err)
+	require.Equal(t, tCtx.GetRayClusterName(), actualCluster.Name)
+	require.Equal(t, rayCluster.Spec, actualCluster.Spec)
+
+	err = rayClient.RayClusters(tCtx.GetNamespaceName()).Delete(tCtx.GetCtx(), tCtx.GetRayClusterName(), metav1.DeleteOptions{})
+	require.NoError(t, err)
+}

apiserver/test/e2e/apiserversdk/constants.go

@@ -0,0 +1,5 @@
+package apiserversdk
+
+const (
+	RayImage = "rayproject/ray:2.9.0"
+)

apiserver/test/e2e/apiserversdk/event_e2e_test.go

@@ -0,0 +1,58 @@
+package apiserversdk
+
+import (
+	"testing"
+
+	"github.com/onsi/gomega"
+	"github.com/stretchr/testify/require"
+	corev1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+
+	"github.com/ray-project/kuberay/apiserver/test/e2e"
+	rayv1 "github.com/ray-project/kuberay/ray-operator/apis/ray/v1"
+)
+
+func TestGetRayClusterEvent(t *testing.T) {
+	tCtx, err := NewEnd2EndTestingContext(t)
+	require.NoError(t, err, "No error expected when creating testing context")
+	rayClient := tCtx.GetRayHttpClient()
+	rayCluster := &rayv1.RayCluster{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      tCtx.GetRayClusterName(),
+			Namespace: tCtx.GetNamespaceName(),
+		},
+		Spec: rayv1.RayClusterSpec{
+			HeadGroupSpec: rayv1.HeadGroupSpec{
+				RayStartParams: map[string]string{},
+				Template: corev1.PodTemplateSpec{
+					Spec: corev1.PodSpec{
+						Containers: []corev1.Container{
+							{Name: "test", Image: "test"},
+						},
+					},
+				},
+			},
+		},
+	}
+	_, err = rayClient.RayClusters(tCtx.GetNamespaceName()).Create(tCtx.GetCtx(), rayCluster, metav1.CreateOptions{})
+	require.NoError(t, err)
+
+	k8sClient := tCtx.GetK8sHttpClient()
+	g := gomega.NewWithT(t)
+	g.Eventually(func() bool {
+		events, err := k8sClient.CoreV1().Events(tCtx.GetNamespaceName()).List(tCtx.GetCtx(), metav1.ListOptions{})
+		if err != nil {
+			return false
+		}
+
+		for _, e := range events.Items {
+			if e.InvolvedObject.Name == tCtx.GetRayClusterName() && e.InvolvedObject.Kind == "RayCluster" {
+				return true
+			}
+		}
+		return false
+	}, e2e.TestTimeoutShort, e2e.TestPollingInterval).Should(gomega.BeTrue(), "Expected to see RayCluster event in the list of events")
+
+	err = rayClient.RayClusters(tCtx.GetNamespaceName()).Delete(tCtx.GetCtx(), tCtx.GetRayClusterName(), metav1.DeleteOptions{})
+	require.NoError(t, err)
+}

apiserver/test/e2e/apiserversdk/proxy_e2e_test.go

@@ -0,0 +1,53 @@
+package apiserversdk
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/require"
+	corev1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+
+	rayv1 "github.com/ray-project/kuberay/ray-operator/apis/ray/v1"
+)
+
+func TestGetRayClusterProxy(t *testing.T) {
+	tCtx, err := NewEnd2EndTestingContext(t)
+	require.NoError(t, err, "No error expected when creating testing context")
+	rayClient := tCtx.GetRayHttpClient()
+	rayCluster := &rayv1.RayCluster{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      tCtx.GetRayClusterName(),
+			Namespace: tCtx.GetNamespaceName(),
+		},
+		Spec: rayv1.RayClusterSpec{
+			HeadGroupSpec: rayv1.HeadGroupSpec{
+				RayStartParams: map[string]string{},
+				Template: corev1.PodTemplateSpec{
+					Spec: corev1.PodSpec{
+						Containers: []corev1.Container{
+							{
+								Name:  "ray-head",
+								Image: tCtx.GetRayImage(),
+							},
+						},
+					},
+				},
+			},
+		},
+	}
+	_, err = rayClient.RayClusters(tCtx.GetNamespaceName()).Create(tCtx.GetCtx(), rayCluster, metav1.CreateOptions{})
+	require.NoError(t, err)
+
+	// Wait for the Ray cluster's head pod to be ready so that we can access the dashboard
+	waitForClusterConditions(t, tCtx, tCtx.GetRayClusterName(), []rayv1.RayClusterConditionType{rayv1.HeadPodReady})
+
+	k8sClient := tCtx.GetK8sHttpClient()
+	serviceName := tCtx.GetRayClusterName() + "-head-svc"
+	r := k8sClient.CoreV1().Services(tCtx.GetNamespaceName()).ProxyGet("http", serviceName, "8265", "", nil)
+	resp, err := r.DoRaw(tCtx.GetCtx())
+	require.NoError(t, err)
+	require.Contains(t, string(resp), "Ray Dashboard")
+
+	err = rayClient.RayClusters(tCtx.GetNamespaceName()).Delete(tCtx.GetCtx(), tCtx.GetRayClusterName(), metav1.DeleteOptions{})
+	require.NoError(t, err)
+}