Merge pull request #93461 from nilo19/automated-cherry-pick-of-#93316…

…-upstream-release-1.17

Automated cherry pick of #93316: Fix instance not found issues when an Azure Node is recreated

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

@@ -174,6 +174,9 @@ type Config struct {
 	// LoadBalancerResourceGroup determines the specific resource group of the load balancer user want to use, working
 	// with LoadBalancerName
 	LoadBalancerResourceGroup string `json:"loadBalancerResourceGroup,omitempty" yaml:"loadBalancerResourceGroup,omitempty"`
+
+	// VmssVirtualMachinesCacheTTLInSeconds sets the cache TTL for vmssVirtualMachines
+	VmssVirtualMachinesCacheTTLInSeconds int `json:"vmssVirtualMachinesCacheTTLInSeconds,omitempty" yaml:"vmssVirtualMachinesCacheTTLInSeconds,omitempty"`
 }
 
 var _ cloudprovider.Interface = (*Cloud)(nil)

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

@@ -256,6 +256,13 @@ func (ss *scaleSet) getVmssVMByInstanceID(resourceGroup, scaleSetName, instanceI
 	if found && vm != nil {
 		return vm, nil
 	}
+	if found && vm == nil {
+		klog.V(2).Infof("Couldn't find VMSS VM with scaleSetName %q and instanceID %q, refreshing the cache if it is expired", scaleSetName, instanceID)
+		vm, found, err = getter(cacheReadTypeDefault)
+		if err != nil {
+			return nil, err
+		}
+	}
 	if !found || vm == nil {
 		return nil, cloudprovider.InstanceNotFound
 	}

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

@@ -42,6 +42,8 @@ var (
 	availabilitySetNodesCacheTTL = 15 * time.Minute
 	vmssTTL                      = 10 * time.Minute
 	vmssVirtualMachinesTTL       = 10 * time.Minute
+
+	vmssVirtualMachinesCacheTTLDefaultInSeconds = 600
 )
 
 type vmssVirtualMachinesEntry struct {
@@ -107,6 +109,11 @@ func extractVmssVMName(name string) (string, string, error) {
 }
 
 func (ss *scaleSet) newVMSSVirtualMachinesCache() (*timedCache, error) {
+	if ss.Config.VmssVirtualMachinesCacheTTLInSeconds == 0 {
+		ss.Config.VmssVirtualMachinesCacheTTLInSeconds = vmssVirtualMachinesCacheTTLDefaultInSeconds
+	}
+	vmssVirtualMachinesCacheTTL := time.Duration(ss.Config.VmssVirtualMachinesCacheTTLInSeconds) * time.Second
+
 	getter := func(key string) (interface{}, error) {
 		localCache := &sync.Map{} // [nodeName]*vmssVirtualMachinesEntry
 
@@ -172,9 +179,9 @@ func (ss *scaleSet) newVMSSVirtualMachinesCache() (*timedCache, error) {
 			// add old missing cache data with nil entries to prevent aggressive
 			// ARM calls during cache invalidation
 			for name, vmEntry := range oldCache {
-				// if the nil cache entry has existed for 15 minutes in the cache
+				// if the nil cache entry has existed for vmssVirtualMachinesCacheTTL in the cache
 				// then it should not be added back to the cache
-				if vmEntry.virtualMachine == nil || time.Since(vmEntry.lastUpdate) > 15*time.Minute {
+				if vmEntry.virtualMachine == nil || time.Since(vmEntry.lastUpdate) > vmssVirtualMachinesCacheTTL {
 					klog.V(5).Infof("ignoring expired entries from old cache for %s", name)
 					continue
 				}