Merge pull request #87635 from feiskyer/cache-nil-for-deleting-node

Set cache to nil data when Azure node provisioning state is deleting

staging/src/k8s.io/legacy-cloud-providers/azure/azure_vmss_cache.go

@@ -156,13 +156,20 @@ func (ss *scaleSet) newVMSSVirtualMachinesCache() (*timedCache, error) {
 					}
 
 					computerName := strings.ToLower(*vm.OsProfile.ComputerName)
-					localCache.Store(computerName, &vmssVirtualMachinesEntry{
+					vmssVMCacheEntry := &vmssVirtualMachinesEntry{
 						resourceGroup:  resourceGroup,
 						vmssName:       ssName,
 						instanceID:     to.String(vm.InstanceID),
 						virtualMachine: &vm,
 						lastUpdate:     time.Now().UTC(),
-					})
+					}
+					// set cache entry to nil when the VM is under deleting.
+					if vm.VirtualMachineScaleSetVMProperties != nil &&
+						strings.EqualFold(to.String(vm.VirtualMachineScaleSetVMProperties.ProvisioningState), string(compute.ProvisioningStateDeleting)) {
+						klog.V(4).Infof("VMSS virtualMachine %q is under deleting, setting its cache to nil", computerName)
+						vmssVMCacheEntry.virtualMachine = nil
+					}
+					localCache.Store(computerName, vmssVMCacheEntry)
 
 					if _, exists := oldCache[computerName]; exists {
 						delete(oldCache, computerName)

staging/src/k8s.io/legacy-cloud-providers/azure/azure_vmss_cache_test.go

@@ -23,8 +23,10 @@ import (
 	"sync"
 	"testing"
 
+	"github.com/Azure/azure-sdk-for-go/services/compute/mgmt/2019-07-01/compute"
 	"github.com/Azure/go-autorest/autorest/to"
 	"github.com/stretchr/testify/assert"
+	cloudprovider "k8s.io/cloud-provider"
 )
 
 func TestExtractVmssVMName(t *testing.T) {
@@ -112,3 +114,26 @@ func TestVMSSVMCache(t *testing.T) {
 	assert.Equal(t, to.String(vm.InstanceID), instanceID)
 	assert.Equal(t, &vm, realVM)
 }
+
+func TestVMSSVMCacheWithDeletingNodes(t *testing.T) {
+	vmssName := "vmss"
+	vmList := []string{"vmssee6c2000000", "vmssee6c2000001", "vmssee6c2000002"}
+	ss, err := newTestScaleSetWithState(vmssName, "", 0, vmList, "Deleting")
+	assert.NoError(t, err)
+
+	virtualMachines, rerr := ss.VirtualMachineScaleSetVMsClient.List(
+		context.Background(), "rg", "vmss", "")
+	assert.Nil(t, rerr)
+	assert.Equal(t, 3, len(virtualMachines))
+	for i := range virtualMachines {
+		vm := virtualMachines[i]
+		vmName := to.String(vm.OsProfile.ComputerName)
+		assert.Equal(t, vm.ProvisioningState, to.StringPtr(string(compute.ProvisioningStateDeleting)))
+
+		ssName, instanceID, realVM, err := ss.getVmssVM(vmName, cacheReadTypeDefault)
+		assert.Nil(t, realVM)
+		assert.Equal(t, "", ssName)
+		assert.Equal(t, instanceID, ssName)
+		assert.Equal(t, cloudprovider.InstanceNotFound, err)
+	}
+}

staging/src/k8s.io/legacy-cloud-providers/azure/azure_vmss_test.go

@@ -34,8 +34,12 @@ const (
 )
 
 func newTestScaleSet(scaleSetName, zone string, faultDomain int32, vmList []string) (*scaleSet, error) {
+	return newTestScaleSetWithState(scaleSetName, zone, faultDomain, vmList, "Running")
+}
+
+func newTestScaleSetWithState(scaleSetName, zone string, faultDomain int32, vmList []string, state string) (*scaleSet, error) {
 	cloud := getTestCloud()
-	setTestVirtualMachineCloud(cloud, scaleSetName, zone, faultDomain, vmList)
+	setTestVirtualMachineCloud(cloud, scaleSetName, zone, faultDomain, vmList, state)
 	ss, err := newScaleSet(cloud)
 	if err != nil {
 		return nil, err
@@ -44,7 +48,7 @@ func newTestScaleSet(scaleSetName, zone string, faultDomain int32, vmList []stri
 	return ss.(*scaleSet), nil
 }
 
-func setTestVirtualMachineCloud(ss *Cloud, scaleSetName, zone string, faultDomain int32, vmList []string) {
+func setTestVirtualMachineCloud(ss *Cloud, scaleSetName, zone string, faultDomain int32, vmList []string, state string) {
 	virtualMachineScaleSetsClient := newFakeVirtualMachineScaleSetsClient()
 	virtualMachineScaleSetVMsClient := newFakeVirtualMachineScaleSetVMsClient()
 	publicIPAddressesClient := newFakeAzurePIPClient("rg")
@@ -99,6 +103,7 @@ func setTestVirtualMachineCloud(ss *Cloud, scaleSetName, zone string, faultDomai
 		}
 		vmssVM := compute.VirtualMachineScaleSetVM{
 			VirtualMachineScaleSetVMProperties: &compute.VirtualMachineScaleSetVMProperties{
+				ProvisioningState: to.StringPtr(state),
 				OsProfile: &compute.OSProfile{
 					ComputerName: &nodeName,
 				},

staging/src/k8s.io/legacy-cloud-providers/azure/azure_wrap.go

@@ -27,6 +27,7 @@ import (
 
 	"github.com/Azure/azure-sdk-for-go/services/compute/mgmt/2019-07-01/compute"
 	"github.com/Azure/azure-sdk-for-go/services/network/mgmt/2019-06-01/network"
+	"github.com/Azure/go-autorest/autorest/to"
 
 	"k8s.io/apimachinery/pkg/types"
 	cloudprovider "k8s.io/cloud-provider"
@@ -193,6 +194,12 @@ func (az *Cloud) newVMCache() (*timedCache, error) {
 			return nil, nil
 		}
 
+		if vm.VirtualMachineProperties != nil &&
+			strings.EqualFold(to.String(vm.VirtualMachineProperties.ProvisioningState), string(compute.ProvisioningStateDeleting)) {
+			klog.V(2).Infof("Virtual machine %q is under deleting", key)
+			return nil, nil
+		}
+
 		return &vm, nil
 	}