Merge pull request #2195 from k8s-infra-cherrypick-robot/cherry-pick-…

…2087-to-release-1.23

[release-1.23] Skip "instance not found" error for LB backend address pools

pkg/provider/azure_loadbalancer.go

@@ -19,6 +19,7 @@ package provider
 import (
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"math"
 	"reflect"
@@ -97,7 +98,9 @@ func (az *Cloud) reconcileService(ctx context.Context, clusterName string, servi
 	lbStatus, fipConfig, err := az.getServiceLoadBalancerStatus(service, lb, nil)
 	if err != nil {
 		klog.Errorf("getServiceLoadBalancerStatus(%s) failed: %v", serviceName, err)
-		return nil, err
+		if !errors.Is(err, ErrorNotVmssInstance) {
+			return nil, err
+		}
 	}
 
 	var serviceIP *string

pkg/provider/azure_vmss.go

@@ -976,7 +976,9 @@ func (ss *ScaleSet) EnsureHostInPool(service *v1.Service, nodeName types.NodeNam
 		}
 
 		klog.Errorf("EnsureHostInPool: failed to get VMSS VM %s: %v", vmName, err)
-		return "", "", "", nil, err
+		if !errors.Is(err, ErrorNotVmssInstance) {
+			return "", "", "", nil, err
+		}
 	}
 
 	klog.V(2).Infof("ensuring node %q of scaleset %q in LB backendpool %q", nodeName, ssName, backendPoolID)