update

Author: ValyaB

internal/actions/patch_node_handler.go

@@ -3,14 +3,14 @@ package actions
 import (
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"reflect"
 	"strconv"
 	"time"
 
 	"github.com/sirupsen/logrus"
 	v1 "k8s.io/api/core/v1"
-	apierrors "k8s.io/apimachinery/pkg/api/errors"
 	"k8s.io/client-go/kubernetes"
 
 	"github.com/castai/cluster-controller/internal/castai"
@@ -21,14 +21,16 @@ var _ ActionHandler = &PatchNodeHandler{}
 
 func NewPatchNodeHandler(log logrus.FieldLogger, clientset kubernetes.Interface) *PatchNodeHandler {
 	return &PatchNodeHandler{
-		log:       log,
-		clientset: clientset,
+		retryTimeout: 5 * time.Second, // default timeout for retrying node patching
+		log:          log,
+		clientset:    clientset,
 	}
 }
 
 type PatchNodeHandler struct {
-	log       logrus.FieldLogger
-	clientset kubernetes.Interface
+	retryTimeout time.Duration
+	log          logrus.FieldLogger
+	clientset    kubernetes.Interface
 }
 
 func (h *PatchNodeHandler) Handle(ctx context.Context, action *castai.ClusterAction) error {
@@ -62,7 +64,7 @@ func (h *PatchNodeHandler) Handle(ctx context.Context, action *castai.ClusterAct
 
 	node, err := h.getNodeForPatching(ctx, req.NodeName, req.NodeID, req.ProviderId)
 	if err != nil {
-		if apierrors.IsNotFound(err) {
+		if errors.Is(err, errNodeNotFound) {
 			log.WithError(err).Infof("node not found, skipping patch")
 			return nil
 		}
@@ -119,7 +121,7 @@ func (h *PatchNodeHandler) getNodeForPatching(ctx context.Context, nodeName, nod
 	var node *v1.Node
 
 	boff := waitext.DefaultExponentialBackoff()
-	boff.Duration = 5 * time.Second
+	boff.Duration = h.retryTimeout
 
 	err := waitext.Retry(
 		ctx,

internal/actions/patch_node_handler_test.go

@@ -3,6 +3,7 @@ package actions
 import (
 	"context"
 	"testing"
+	"time"
 
 	"github.com/google/uuid"
 	"github.com/samber/lo"
@@ -24,6 +25,7 @@ func TestPatchNodeHandler(t *testing.T) {
 
 	t.Run("patch successfully", func(t *testing.T) {
 		nodeName := "node1"
+		providerID := "provider-id-123"
 		node := &v1.Node{
 			ObjectMeta: metav1.ObjectMeta{
 				Name: nodeName,
@@ -35,6 +37,7 @@ func TestPatchNodeHandler(t *testing.T) {
 				},
 			},
 			Spec: v1.NodeSpec{
+				ProviderID: providerID,
 				Taints: []v1.Taint{
 					{
 						Key:    "t1",
@@ -59,7 +62,8 @@ func TestPatchNodeHandler(t *testing.T) {
 		action := &castai.ClusterAction{
 			ID: uuid.New().String(),
 			ActionPatchNode: &castai.ActionPatchNode{
-				NodeName: "node1",
+				NodeName:   "node1",
+				ProviderId: providerID,
 				Labels: map[string]string{
 					"-l1": "",
 					"l2":  "v2",
@@ -113,6 +117,7 @@ func TestPatchNodeHandler(t *testing.T) {
 
 	t.Run("skip patch when node not found", func(t *testing.T) {
 		nodeName := "node1"
+		nodeID := "node-id-123"
 		node := &v1.Node{
 			ObjectMeta: metav1.ObjectMeta{
 				Name: nodeName,
@@ -124,11 +129,13 @@ func TestPatchNodeHandler(t *testing.T) {
 			ID: uuid.New().String(),
 			ActionPatchNode: &castai.ActionPatchNode{
 				NodeName: "already-deleted-node",
+				NodeID:   nodeID,
 			},
 		}
 		h := PatchNodeHandler{
-			log:       log,
-			clientset: clientset,
+			retryTimeout: time.Millisecond,
+			log:          log,
+			clientset:    clientset,
 		}
 
 		err := h.Handle(context.Background(), action)
@@ -140,9 +147,13 @@ func TestPatchNodeHandler(t *testing.T) {
 
 	t.Run("cordoning node", func(t *testing.T) {
 		nodeName := "node1"
+		nodeID := "node-id-123"
 		node := &v1.Node{
 			ObjectMeta: metav1.ObjectMeta{
 				Name: nodeName,
+				Labels: map[string]string{
+					castai.LabelNodeID: nodeID,
+				},
 			},
 			Spec: v1.NodeSpec{
 				Unschedulable: false,
@@ -159,6 +170,7 @@ func TestPatchNodeHandler(t *testing.T) {
 			ID: uuid.New().String(),
 			ActionPatchNode: &castai.ActionPatchNode{
 				NodeName:      "node1",
+				NodeID:        nodeID,
 				Unschedulable: lo.ToPtr(true),
 			},
 		}