added state restoring from checkpoint file

Change-Id: I52072035a80f2cff6b781d530f8a8381b7c81d66

checkpoint/checkpoint.go

@@ -0,0 +1,60 @@
+package checkpoint
+
+import (
+	"encoding/json"
+	"fmt"
+	"io/ioutil"
+
+	pluginapi "k8s.io/kubernetes/pkg/kubelet/apis/deviceplugin/v1beta1"
+	"k8s.io/kubernetes/pkg/kubelet/checkpointmanager/checksum"
+)
+
+const (
+	checkPointfile                 = "/var/lib/kubelet/device-plugins/kubelet_internal_checkpoint"
+	pluginDir                      = pluginapi.DevicePluginPath
+	kubeletDeviceManagerCheckpoint = "kubelet_internal_checkpoint"
+)
+
+type PodDevicesEntry struct {
+	PodUID        string
+	ContainerName string
+	ResourceName  string
+	DeviceIDs     []string
+	AllocResp     []byte
+}
+
+type checkpointData struct {
+	PodDeviceEntries  []PodDevicesEntry
+	RegisteredDevices map[string][]string
+}
+
+type Data struct {
+	Data     checkpointData
+	Checksum checksum.Checksum
+}
+
+// Get all Pod entires for given resourceName
+func GetPodEntries(resourceName string) ([]PodDevicesEntry, error) {
+
+	podEntries := []PodDevicesEntry{}
+
+	cpd := &Data{}
+	rawBytes, err := ioutil.ReadFile(checkPointfile)
+	if err != nil {
+		return podEntries, fmt.Errorf("Error reading file %s\n%v\n", checkPointfile, err)
+
+	}
+
+	if err = json.Unmarshal(rawBytes, cpd); err != nil {
+		return podEntries, fmt.Errorf("Error unmarshalling raw bytes %v", err)
+	}
+
+	responseData := cpd.Data.PodDeviceEntries
+
+	for _, p := range responseData {
+		if p.ResourceName == resourceName {
+			podEntries = append(podEntries, p)
+		}
+	}
+	return podEntries, nil
+}

cmd/sriovdp/sriov-device-plugin.go

@@ -33,6 +33,7 @@ import (
 	"google.golang.org/grpc"
 
 	"github.com/intel/sriov-network-device-plugin/api"
+	"github.com/intel/sriov-network-device-plugin/checkpoint"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/client-go/kubernetes"
 	"k8s.io/client-go/rest"
@@ -79,13 +80,37 @@ func newSriovManager() *sriovManager {
 		glog.Errorf("Error. Could not create K8s Client using supplied config. %v", err)
 		return nil
 	}
-	return &sriovManager{
+	sm := &sriovManager{
 		k8ClientSet:      clientset,
 		devices:          make(map[string]pluginapi.Device),
 		allocatedDevices: make(map[string][]*deviceEntry),
 		managedDevices:   make(map[string]*api.VfInformation),
 		socketFile:       fmt.Sprintf("%s.sock", pluginEndpointPrefix),
 	}
+	// get device mapping from checkpoint if there's any
+	restorePodMapping(sm)
+
+	return sm
+}
+
+// Restore Pod to device mapping from checkpoint file
+func restorePodMapping(sm *sriovManager) {
+	glog.Infof("Checking checkpoint file")
+	podEntries, err := checkpoint.GetPodEntries(resourceName)
+	if err == nil && len(podEntries) > 0 {
+		for _, p := range podEntries {
+			logStr := fmt.Sprintf("Restored PodUID: %s DeviceIDs : {", p.PodUID)
+			dEntry := []*deviceEntry{}
+			for _, d := range p.DeviceIDs {
+				// assuming CNI plugin already configured these VFs mark them as configured 'true'
+				dEntry = append(dEntry, &deviceEntry{deviceID: d, allocated: true})
+				logStr += fmt.Sprintf("%s,", d)
+			}
+			logStr += "}"
+			sm.allocatedDevices[p.PodUID] = dEntry
+			glog.Infof(logStr)
+		}
+	}
 }
 
 // Returns a list of SRIOV capable PF names as string
@@ -315,6 +340,7 @@ func (sm *sriovManager) ListAndWatch(empty *pluginapi.Empty, stream pluginapi.De
 		changed = false
 		time.Sleep(5 * time.Second)
 	}
+	return nil
 }
 
 func (sm *sriovManager) PreStartContainer(ctx context.Context, psRqt *pluginapi.PreStartContainerRequest) (*pluginapi.PreStartContainerResponse, error) {
@@ -427,8 +453,12 @@ func (sm *sriovManager) CNIDelete(ctx context.Context, podInfo *api.PodInformati
 
 func main() {
 	flag.Parse()
-	glog.Infof("SRIOV Network Device Plugin started...")
+	glog.Infof("Starting SRIOV Network Device Plugin...")
 	sm := newSriovManager()
+	if sm == nil {
+		glog.Errorf("Unable to get instance of a SRIOV-Manager")
+		return
+	}
 	sm.cleanup()
 
 	// respond to syscalls for termination