feat(zrpc): migrate kube resolver from Endpoints to EndpointSlice API (#4987)

Signed-off-by: soasurs <soasurs@gmail.com>
Co-authored-by: Kevin Wan <wanjunfeng@gmail.com>

.github/workflows/reviewdog.yml

@@ -6,6 +6,11 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v6
+      - uses: actions/setup-go@v6
+        with:
+          go-version-file: go.mod
+          check-latest: true
+          cache: true
       - uses: reviewdog/action-staticcheck@v1
         with:
           github_token: ${{ secrets.github_token }}

zrpc/resolver/internal/kube/eventhandler.go

@@ -5,7 +5,7 @@ import (
 
 	"github.com/zeromicro/go-zero/core/lang"
 	"github.com/zeromicro/go-zero/core/logx"
-	"k8s.io/api/core/v1"
+	"k8s.io/api/discovery/v1"
 	"k8s.io/client-go/tools/cache"
 )
 
@@ -28,9 +28,9 @@ func NewEventHandler(update func([]string)) *EventHandler {
 
 // OnAdd handles the endpoints add events.
 func (h *EventHandler) OnAdd(obj any, _ bool) {
-	endpoints, ok := obj.(*v1.Endpoints)
+	endpoints, ok := obj.(*v1.EndpointSlice)
 	if !ok {
-		logx.Errorf("%v is not an object with type *v1.Endpoints", obj)
+		logx.Errorf("%v is not an object with type *v1.EndpointSlice", obj)
 		return
 	}
 
@@ -38,10 +38,10 @@ func (h *EventHandler) OnAdd(obj any, _ bool) {
 	defer h.lock.Unlock()
 
 	var changed bool
-	for _, sub := range endpoints.Subsets {
+	for _, point := range endpoints.Endpoints {
-		for _, point := range sub.Addresses {
+		for _, address := range point.Addresses {
-			if _, ok := h.endpoints[point.IP]; !ok {
+			if _, ok := h.endpoints[address]; !ok {
-				h.endpoints[point.IP] = lang.Placeholder
+				h.endpoints[address] = lang.Placeholder
 				changed = true
 			}
 		}
@@ -54,9 +54,9 @@ func (h *EventHandler) OnAdd(obj any, _ bool) {
 
 // OnDelete handles the endpoints delete events.
 func (h *EventHandler) OnDelete(obj any) {
-	endpoints, ok := obj.(*v1.Endpoints)
+	endpoints, ok := obj.(*v1.EndpointSlice)
 	if !ok {
-		logx.Errorf("%v is not an object with type *v1.Endpoints", obj)
+		logx.Errorf("%v is not an object with type *v1.EndpointSlice", obj)
 		return
 	}
 
@@ -64,10 +64,10 @@ func (h *EventHandler) OnDelete(obj any) {
 	defer h.lock.Unlock()
 
 	var changed bool
-	for _, sub := range endpoints.Subsets {
+	for _, point := range endpoints.Endpoints {
-		for _, point := range sub.Addresses {
+		for _, address := range point.Addresses {
-			if _, ok := h.endpoints[point.IP]; ok {
+			if _, ok := h.endpoints[address]; ok {
-				delete(h.endpoints, point.IP)
+				delete(h.endpoints, address)
 				changed = true
 			}
 		}
@@ -80,35 +80,35 @@ func (h *EventHandler) OnDelete(obj any) {
 
 // OnUpdate handles the endpoints update events.
 func (h *EventHandler) OnUpdate(oldObj, newObj any) {
-	oldEndpoints, ok := oldObj.(*v1.Endpoints)
+	oldEndpointSlices, ok := oldObj.(*v1.EndpointSlice)
 	if !ok {
-		logx.Errorf("%v is not an object with type *v1.Endpoints", oldObj)
+		logx.Errorf("%v is not an object with type *v1.EndpointSlice", oldObj)
 		return
 	}
 
-	newEndpoints, ok := newObj.(*v1.Endpoints)
+	newEndpointSlices, ok := newObj.(*v1.EndpointSlice)
 	if !ok {
-		logx.Errorf("%v is not an object with type *v1.Endpoints", newObj)
+		logx.Errorf("%v is not an object with type *v1.EndpointSlice", newObj)
 		return
 	}
 
-	if oldEndpoints.ResourceVersion == newEndpoints.ResourceVersion {
+	if oldEndpointSlices.ResourceVersion == newEndpointSlices.ResourceVersion {
 		return
 	}
 
-	h.Update(newEndpoints)
+	h.Update(newEndpointSlices)
 }
 
 // Update updates the endpoints.
-func (h *EventHandler) Update(endpoints *v1.Endpoints) {
+func (h *EventHandler) Update(endpoints *v1.EndpointSlice) {
 	h.lock.Lock()
 	defer h.lock.Unlock()
 
 	old := h.endpoints
 	h.endpoints = make(map[string]lang.PlaceholderType)
-	for _, sub := range endpoints.Subsets {
+	for _, point := range endpoints.Endpoints {
-		for _, point := range sub.Addresses {
+		for _, address := range point.Addresses {
-			h.endpoints[point.IP] = lang.Placeholder
+			h.endpoints[address] = lang.Placeholder
 		}
 	}
 

zrpc/resolver/internal/kube/eventhandler_test.go

@@ -4,7 +4,7 @@ import (
 	"testing"
 
 	"github.com/stretchr/testify/assert"
-	v1 "k8s.io/api/core/v1"
+	discoveryv1 "k8s.io/api/discovery/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 )
 
@@ -14,21 +14,19 @@ func TestAdd(t *testing.T) {
 		endpoints = change
 	})
 	h.OnAdd("bad", false)
-	h.OnAdd(&v1.Endpoints{Subsets: []v1.EndpointSubset{
+	h.OnAdd(&discoveryv1.EndpointSlice{
+		Endpoints: []discoveryv1.Endpoint{
 			{
-			Addresses: []v1.EndpointAddress{
+				Addresses: []string{"0.0.0.1"},
-				{
-					IP: "0.0.0.1",
 			},
 			{
-					IP: "0.0.0.2",
+				Addresses: []string{"0.0.0.2"},
 			},
 			{
-					IP: "0.0.0.3",
+				Addresses: []string{"0.0.0.3"},
 			},
 		},
-		},
+	}, false)
-	}}, false)
 	assert.ElementsMatch(t, []string{"0.0.0.1", "0.0.0.2", "0.0.0.3"}, endpoints)
 }
 
@@ -37,34 +35,30 @@ func TestDelete(t *testing.T) {
 	h := NewEventHandler(func(change []string) {
 		endpoints = change
 	})
-	h.OnAdd(&v1.Endpoints{Subsets: []v1.EndpointSubset{
+	h.OnAdd(&discoveryv1.EndpointSlice{
+		Endpoints: []discoveryv1.Endpoint{
 			{
-			Addresses: []v1.EndpointAddress{
+				Addresses: []string{"0.0.0.1"},
-				{
-					IP: "0.0.0.1",
 			},
 			{
-					IP: "0.0.0.2",
+				Addresses: []string{"0.0.0.2"},
 			},
 			{
-					IP: "0.0.0.3",
+				Addresses: []string{"0.0.0.3"},
 			},
 		},
-		},
+	}, false)
-	}}, false)
 	h.OnDelete("bad")
-	h.OnDelete(&v1.Endpoints{Subsets: []v1.EndpointSubset{
+	h.OnDelete(&discoveryv1.EndpointSlice{
+		Endpoints: []discoveryv1.Endpoint{
 			{
-			Addresses: []v1.EndpointAddress{
+				Addresses: []string{"0.0.0.1"},
-				{
-					IP: "0.0.0.1",
 			},
 			{
-					IP: "0.0.0.2",
+				Addresses: []string{"0.0.0.2"},
 			},
 		},
-		},
+	})
-	}})
 	assert.ElementsMatch(t, []string{"0.0.0.3"}, endpoints)
 }
 
@@ -73,36 +67,28 @@ func TestUpdate(t *testing.T) {
 	h := NewEventHandler(func(change []string) {
 		endpoints = change
 	})
-	h.OnUpdate(&v1.Endpoints{
+	h.OnUpdate(&discoveryv1.EndpointSlice{
-		Subsets: []v1.EndpointSubset{
+		Endpoints: []discoveryv1.Endpoint{
 			{
-				Addresses: []v1.EndpointAddress{
+				Addresses: []string{"0.0.0.1"},
-					{
-						IP: "0.0.0.1",
 			},
 			{
-						IP: "0.0.0.2",
+				Addresses: []string{"0.0.0.2"},
-					},
-				},
 			},
 		},
 		ObjectMeta: metav1.ObjectMeta{
 			ResourceVersion: "1",
 		},
-	}, &v1.Endpoints{
+	}, &discoveryv1.EndpointSlice{
-		Subsets: []v1.EndpointSubset{
+		Endpoints: []discoveryv1.Endpoint{
 			{
-				Addresses: []v1.EndpointAddress{
+				Addresses: []string{"0.0.0.1"},
-					{
-						IP: "0.0.0.1",
 			},
 			{
-						IP: "0.0.0.2",
+				Addresses: []string{"0.0.0.2"},
 			},
 			{
-						IP: "0.0.0.3",
+				Addresses: []string{"0.0.0.3"},
-					},
-				},
 			},
 		},
 		ObjectMeta: metav1.ObjectMeta{
@@ -116,33 +102,25 @@ func TestUpdateNoChange(t *testing.T) {
 	h := NewEventHandler(func(change []string) {
 		assert.Fail(t, "should not called")
 	})
-	h.OnUpdate(&v1.Endpoints{
+	h.OnUpdate(&discoveryv1.EndpointSlice{
-		Subsets: []v1.EndpointSubset{
+		Endpoints: []discoveryv1.Endpoint{
 			{
-				Addresses: []v1.EndpointAddress{
+				Addresses: []string{"0.0.0.1"},
-					{
-						IP: "0.0.0.1",
 			},
 			{
-						IP: "0.0.0.2",
+				Addresses: []string{"0.0.0.2"},
-					},
-				},
 			},
 		},
 		ObjectMeta: metav1.ObjectMeta{
 			ResourceVersion: "1",
 		},
-	}, &v1.Endpoints{
+	}, &discoveryv1.EndpointSlice{
-		Subsets: []v1.EndpointSubset{
+		Endpoints: []discoveryv1.Endpoint{
 			{
-				Addresses: []v1.EndpointAddress{
+				Addresses: []string{"0.0.0.1"},
-					{
-						IP: "0.0.0.1",
 			},
 			{
-						IP: "0.0.0.2",
+				Addresses: []string{"0.0.0.2"},
-					},
-				},
 			},
 		},
 		ObjectMeta: metav1.ObjectMeta{
@@ -156,45 +134,35 @@ func TestUpdateChangeWithDifferentVersion(t *testing.T) {
 	h := NewEventHandler(func(change []string) {
 		endpoints = change
 	})
-	h.OnAdd(&v1.Endpoints{Subsets: []v1.EndpointSubset{
+	h.OnAdd(&discoveryv1.EndpointSlice{
+		Endpoints: []discoveryv1.Endpoint{
 			{
-			Addresses: []v1.EndpointAddress{
+				Addresses: []string{"0.0.0.1"},
-				{
-					IP: "0.0.0.1",
 			},
 			{
-					IP: "0.0.0.3",
+				Addresses: []string{"0.0.0.3"},
 			},
 		},
-		},
+	}, false)
-	}}, false)
+	h.OnUpdate(&discoveryv1.EndpointSlice{
-	h.OnUpdate(&v1.Endpoints{
+		Endpoints: []discoveryv1.Endpoint{
-		Subsets: []v1.EndpointSubset{
 			{
-				Addresses: []v1.EndpointAddress{
+				Addresses: []string{"0.0.0.1"},
-					{
-						IP: "0.0.0.1",
 			},
 			{
-						IP: "0.0.0.3",
+				Addresses: []string{"0.0.0.3"},
-					},
-				},
 			},
 		},
 		ObjectMeta: metav1.ObjectMeta{
 			ResourceVersion: "1",
 		},
-	}, &v1.Endpoints{
+	}, &discoveryv1.EndpointSlice{
-		Subsets: []v1.EndpointSubset{
+		Endpoints: []discoveryv1.Endpoint{
 			{
-				Addresses: []v1.EndpointAddress{
+				Addresses: []string{"0.0.0.1"},
-					{
-						IP: "0.0.0.1",
 			},
 			{
-						IP: "0.0.0.2",
+				Addresses: []string{"0.0.0.2"},
-					},
-				},
 			},
 		},
 		ObjectMeta: metav1.ObjectMeta{
@@ -209,63 +177,49 @@ func TestUpdateNoChangeWithDifferentVersion(t *testing.T) {
 	h := NewEventHandler(func(change []string) {
 		endpoints = change
 	})
-	h.OnAdd(&v1.Endpoints{Subsets: []v1.EndpointSubset{
+	h.OnAdd(&discoveryv1.EndpointSlice{
+		Endpoints: []discoveryv1.Endpoint{
 			{
-			Addresses: []v1.EndpointAddress{
+				Addresses: []string{"0.0.0.1"},
-				{
-					IP: "0.0.0.1",
 			},
 			{
-					IP: "0.0.0.2",
+				Addresses: []string{"0.0.0.2"},
 			},
 		},
-		},
+	}, false)
-	}}, false)
+	h.OnUpdate("bad", &discoveryv1.EndpointSlice{
-	h.OnUpdate("bad", &v1.Endpoints{Subsets: []v1.EndpointSubset{
+		Endpoints: []discoveryv1.Endpoint{
 			{
-			Addresses: []v1.EndpointAddress{
+				Addresses: []string{"0.0.0.1"},
+			},
+		},
+	})
+	h.OnUpdate(&discoveryv1.EndpointSlice{
+		Endpoints: []discoveryv1.Endpoint{
 			{
-					IP: "0.0.0.1",
+				Addresses: []string{"0.0.0.1"},
 			},
 		},
-		},
+	}, "bad")
-	}})
+	h.OnUpdate(&discoveryv1.EndpointSlice{
-	h.OnUpdate(&v1.Endpoints{Subsets: []v1.EndpointSubset{
+		Endpoints: []discoveryv1.Endpoint{
 			{
-			Addresses: []v1.EndpointAddress{
+				Addresses: []string{"0.0.0.1"},
-				{
-					IP: "0.0.0.1",
-				},
-			},
-		},
-	}}, "bad")
-	h.OnUpdate(&v1.Endpoints{
-		Subsets: []v1.EndpointSubset{
-			{
-				Addresses: []v1.EndpointAddress{
-					{
-						IP: "0.0.0.1",
 			},
 			{
-						IP: "0.0.0.2",
+				Addresses: []string{"0.0.0.2"},
-					},
-				},
 			},
 		},
 		ObjectMeta: metav1.ObjectMeta{
 			ResourceVersion: "1",
 		},
-	}, &v1.Endpoints{
+	}, &discoveryv1.EndpointSlice{
-		Subsets: []v1.EndpointSubset{
+		Endpoints: []discoveryv1.Endpoint{
 			{
-				Addresses: []v1.EndpointAddress{
+				Addresses: []string{"0.0.0.1"},
-					{
-						IP: "0.0.0.1",
 			},
 			{
-						IP: "0.0.0.2",
+				Addresses: []string{"0.0.0.2"},
-					},
-				},
 			},
 		},
 		ObjectMeta: metav1.ObjectMeta{

zrpc/resolver/internal/kubebuilder.go

@@ -19,7 +19,7 @@ import (
 
 const (
 	resyncInterval  = 5 * time.Minute
-	nameSelector   = "metadata.name="
+	serviceSelector = "kubernetes.io/service-name="
 )
 
 type kubeResolver struct {
@@ -60,14 +60,33 @@ func (b *kubeBuilder) Build(target resolver.Target, cc resolver.ClientConn,
 	}
 
 	if svc.Port == 0 {
-		// getting endpoints is only to get the port
+		endpointSlices, err := cs.DiscoveryV1().EndpointSlices(svc.Namespace).List(context.Background(),
-		endpoints, err := cs.CoreV1().Endpoints(svc.Namespace).Get(
+			v1.ListOptions{
-			context.Background(), svc.Name, v1.GetOptions{})
+				LabelSelector: serviceSelector + svc.Name,
+			})
 		if err != nil {
 			return nil, err
 		}
+		if len(endpointSlices.Items) == 0 {
+			return nil, fmt.Errorf("no endpoint slices found for service %s in namespace %s",
+				svc.Name, svc.Namespace)
+		}
 
-		svc.Port = int(endpoints.Subsets[0].Ports[0].Port)
+		// Find the first slice with a valid port.
+		// Since this resolver is used for in-cluster service discovery,
+		// we expect at least one port to be available.
+		var foundPort bool
+		for _, slice := range endpointSlices.Items {
+			if len(slice.Ports) > 0 && slice.Ports[0].Port != nil {
+				svc.Port = int(*slice.Ports[0].Port)
+				foundPort = true
+				break
+			}
+		}
+		if !foundPort {
+			return nil, fmt.Errorf("no valid port found in endpoint slices for service %s in namespace %s",
+				svc.Name, svc.Namespace)
+		}
 	}
 
 	handler := kube.NewEventHandler(func(endpoints []string) {
@@ -88,23 +107,29 @@ func (b *kubeBuilder) Build(target resolver.Target, cc resolver.ClientConn,
 	inf := informers.NewSharedInformerFactoryWithOptions(cs, resyncInterval,
 		informers.WithNamespace(svc.Namespace),
 		informers.WithTweakListOptions(func(options *v1.ListOptions) {
-			options.FieldSelector = nameSelector + svc.Name
+			options.LabelSelector = serviceSelector + svc.Name
 		}))
-	in := inf.Core().V1().Endpoints()
+	in := inf.Discovery().V1().EndpointSlices()
 	_, err = in.Informer().AddEventHandler(handler)
 	if err != nil {
 		return nil, err
 	}
 
-	// get the initial endpoints, cannot use the previous endpoints,
+	// get the initial endpoint slices, cannot use the previous endpoint slices,
-	// because the endpoints may be updated before/after the informer is started.
+	// because the endpoint slices may be updated before/after the informer is started.
-	endpoints, err := cs.CoreV1().Endpoints(svc.Namespace).Get(
+	endpointSlices, err := cs.DiscoveryV1().EndpointSlices(svc.Namespace).List(
-		context.Background(), svc.Name, v1.GetOptions{})
+		context.Background(), v1.ListOptions{
+			LabelSelector: serviceSelector + svc.Name,
+		})
 	if err != nil {
 		return nil, err
 	}
 
-	handler.Update(endpoints)
+	// Aggregate endpoints from all EndpointSlices.
+	// Use OnAdd (not Update) to accumulate addresses across multiple slices.
+	for _, endpointSlice := range endpointSlices.Items {
+		handler.OnAdd(&endpointSlice, false)
+	}
 
 	r := &kubeResolver{
 		cc:     cc,