Add cache for virtual machines

pkg/cloudprovider/providers/azure/azure.go

@@ -135,6 +135,8 @@ type Cloud struct {
 	VirtualMachineScaleSetsClient   VirtualMachineScaleSetsClient
 	VirtualMachineScaleSetVMsClient VirtualMachineScaleSetVMsClient
 
+	vmCache *timedCache
+
 	*BlobDiskController
 	*ManagedDiskController
 	*controllerCommon
@@ -244,6 +246,12 @@ func NewCloud(configReader io.Reader) (cloudprovider.Interface, error) {
 		az.vmSet = newAvailabilitySet(&az)
 	}
 
+	vmCache, err := az.newVMCache()
+	if err != nil {
+		return nil, err
+	}
+	az.vmCache = vmCache
+
 	if err := initDiskControllers(&az); err != nil {
 		return nil, err
 	}

pkg/cloudprovider/providers/azure/azure_controllerCommon.go

@@ -124,7 +124,7 @@ func (c *controllerCommon) AttachDisk(isManagedDisk bool, diskName, diskURI stri
 	} else {
 		glog.V(4).Info("azureDisk - azure attach succeeded")
 		// Invalidate the cache right after updating
-		vmCache.Delete(vmName)
+		c.cloud.vmCache.Delete(vmName)
 	}
 	return err
 }
@@ -183,7 +183,7 @@ func (c *controllerCommon) DetachDiskByName(diskName, diskURI string, nodeName t
 	} else {
 		glog.V(4).Info("azureDisk - azure disk detach succeeded")
 		// Invalidate the cache right after updating
-		vmCache.Delete(vmName)
+		c.cloud.vmCache.Delete(vmName)
 	}
 	return err
 }

pkg/cloudprovider/providers/azure/azure_test.go

@@ -878,6 +878,7 @@ func getTestCloud() (az *Cloud) {
 	az.VirtualMachineScaleSetVMsClient = newFakeVirtualMachineScaleSetVMsClient()
 	az.VirtualMachinesClient = newFakeAzureVirtualMachinesClient()
 	az.vmSet = newAvailabilitySet(az)
+	az.vmCache, _ = az.newVMCache()
 
 	return az
 }

pkg/cloudprovider/providers/azure/azure_wrap.go

@@ -18,18 +18,20 @@ package azure
 
 import (
 	"net/http"
-	"sync"
 	"time"
 
 	"github.com/Azure/azure-sdk-for-go/arm/compute"
 	"github.com/Azure/azure-sdk-for-go/arm/network"
 	"github.com/Azure/go-autorest/autorest"
-	"github.com/golang/glog"
 
 	"k8s.io/apimachinery/pkg/types"
 	"k8s.io/kubernetes/pkg/cloudprovider"
 )
 
+var (
+	vmCacheTTL = time.Minute
+)
+
 // checkExistsFromError inspects an error and returns a true if err is nil,
 // false if error is an autorest.Error with StatusCode=404 and will return the
 // error back if error is another status code or another type of error.
@@ -60,59 +62,21 @@ func ignoreStatusNotFoundFromError(err error) error {
 	return err
 }
 
-// cache used by getVirtualMachine
-// 15s for expiration duration
-var vmCache = newTimedcache(15 * time.Second)
-
-type vmRequest struct {
-	lock *sync.Mutex
-	vm   *compute.VirtualMachine
-}
-
 /// getVirtualMachine calls 'VirtualMachinesClient.Get' with a timed cache
 /// The service side has throttling control that delays responses if there're multiple requests onto certain vm
 /// resource request in short period.
 func (az *Cloud) getVirtualMachine(nodeName types.NodeName) (vm compute.VirtualMachine, err error) {
 	vmName := string(nodeName)
-
-	cachedRequest, err := vmCache.GetOrCreate(vmName, func() interface{} {
-		return &vmRequest{
-			lock: &sync.Mutex{},
-			vm:   nil,
-		}
-	})
+	cachedVM, err := az.vmCache.Get(vmName)
 	if err != nil {
-		return compute.VirtualMachine{}, err
-	}
-	request := cachedRequest.(*vmRequest)
-
-	if request.vm == nil {
-		request.lock.Lock()
-		defer request.lock.Unlock()
-		if request.vm == nil {
-			// Currently InstanceView request are used by azure_zones, while the calls come after non-InstanceView
-			// request. If we first send an InstanceView request and then a non InstanceView request, the second
-			// request will still hit throttling. This is what happens now for cloud controller manager: In this
-			// case we do get instance view every time to fulfill the azure_zones requirement without hitting
-			// throttling.
-			// Consider adding separate parameter for controlling 'InstanceView' once node update issue #56276 is fixed
-			vm, err = az.VirtualMachinesClient.Get(az.ResourceGroup, vmName, compute.InstanceView)
-			exists, realErr := checkResourceExistsFromError(err)
-			if realErr != nil {
-				return vm, realErr
-			}
-
-			if !exists {
-				return vm, cloudprovider.InstanceNotFound
-			}
-
-			request.vm = &vm
-		}
-		return *request.vm, nil
+		return vm, err
+	}
+
+	if cachedVM == nil {
+		return vm, cloudprovider.InstanceNotFound
 	}
 
-	glog.V(6).Infof("getVirtualMachine hits cache for(%s)", vmName)
-	return *request.vm, nil
+	return *(cachedVM.(*compute.VirtualMachine)), nil
 }
 
 func (az *Cloud) getRouteTable() (routeTable network.RouteTable, exists bool, err error) {
@@ -189,3 +153,21 @@ func (az *Cloud) getAzureLoadBalancer(name string) (lb network.LoadBalancer, exi
 
 	return lb, exists, err
 }
+
+func (az *Cloud) newVMCache() (*timedCache, error) {
+	getter := func(key string) (interface{}, error) {
+		vm, err := az.VirtualMachinesClient.Get(az.ResourceGroup, key, compute.InstanceView)
+		exists, realErr := checkResourceExistsFromError(err)
+		if realErr != nil {
+			return nil, realErr
+		}
+
+		if !exists {
+			return nil, nil
+		}
+
+		return &vm, nil
+	}
+
+	return newTimedcache(vmCacheTTL, getter)
+}