fix: when multi container use device allocate fail

Signed-off-by: rongfu.leng <[email protected]>

pkg/device-plugin/hygon/dcu/server.go

@@ -547,7 +547,7 @@ func (p *Plugin) Allocate(ctx context.Context, reqs *pluginapi.AllocateRequest)
 		responses.ContainerResponses = append(responses.ContainerResponses, &car)
 	}
 	klog.Infoln("response=", responses)
-	device.PodAllocationTrySuccess(nodename, current)
+	device.PodAllocationTrySuccess(nodename, hygon.HygonDCUDevice, current)
 	return &responses, nil
 }
 

pkg/device-plugin/mlu/server.go

@@ -334,7 +334,7 @@ func (m *CambriconDevicePlugin) allocateMLUShare(ctx context.Context, reqs *plug
 		responses.ContainerResponses = append(responses.ContainerResponses, &resp)
 	}
 	klog.Infoln("response=", responses)
-	device.PodAllocationTrySuccess(nodename, current)
+	device.PodAllocationTrySuccess(nodename, cambricon.CambriconMLUDevice, current)
 	return &responses, nil
 }
 

pkg/device-plugin/nvidiadevice/nvinternal/plugin/server.go

@@ -294,6 +294,7 @@ func (plugin *NvidiaDevicePlugin) Allocate(ctx context.Context, reqs *pluginapi.
 		nodelock.ReleaseNodeLock(nodename)
 		return &pluginapi.AllocateResponse{}, err
 	}
+	klog.V(5).Infof("allocate pod name is %s/%s, annotation is %+v", current.Namespace, current.Name, current.Annotations)
 
 	for idx, req := range reqs.ContainerRequests {
 		// If the devices being allocated are replicas, then (conditionally)
@@ -406,7 +407,7 @@ func (plugin *NvidiaDevicePlugin) Allocate(ctx context.Context, reqs *pluginapi.
 		}
 	}
 	klog.Infoln("Allocate Response", responses.ContainerResponses)
-	device.PodAllocationTrySuccess(nodename, current)
+	device.PodAllocationTrySuccess(nodename, nvidia.NvidiaGPUDevice, current)
 	return &responses, nil
 }
 

pkg/device/devices.go

@@ -56,9 +56,13 @@ func init() {
 	DevicesToHandle = append(DevicesToHandle, iluvatar.IluvatarGPUCommonWord)
 }
 
-func PodAllocationTrySuccess(nodeName string, pod *v1.Pod) {
-	refreshed, _ := client.GetClient().CoreV1().Pods(pod.Namespace).Get(context.Background(), pod.Name, metav1.GetOptions{})
-	annos := refreshed.Annotations[util.AssignedIDsToAllocateAnnotations]
+func PodAllocationTrySuccess(nodeName string, devName string, pod *v1.Pod) {
+	refreshed, err := client.GetClient().CoreV1().Pods(pod.Namespace).Get(context.Background(), pod.Name, metav1.GetOptions{})
+	if err != nil {
+		klog.Errorf("get pods %s/%s error: %+v", pod.Namespace, pod.Name, err)
+		return
+	}
+	annos := refreshed.Annotations[util.InRequestDevices[devName]]
 	klog.Infoln("TrySuccess:", annos)
 	for _, val := range DevicesToHandle {
 		if strings.Contains(annos, val) {

pkg/device/nvidia/device.go

@@ -165,10 +165,11 @@ func (dev *NvidiaGPUDevices) CheckType(annos map[string]string, d util.DeviceUsa
 func (dev *NvidiaGPUDevices) PatchAnnotations(annoinput *map[string]string, pd util.PodDevices) map[string]string {
 	devlist, ok := pd[NvidiaGPUDevice]
 	if ok && len(devlist) > 0 {
-		(*annoinput)[util.InRequestDevices[NvidiaGPUDevice]] = util.EncodePodSingleDevice(devlist)
-		(*annoinput)[util.SupportDevices[NvidiaGPUDevice]] = util.EncodePodSingleDevice(devlist)
-		//InRequestDevices := util.EncodePodDevices(util.InRequestDevices, m.devices)
-		//supportDevices := util.EncodePodDevices(util.SupportDevices, m.devices)
+		deviceStr := util.EncodePodSingleDevice(devlist)
+		(*annoinput)[util.InRequestDevices[NvidiaGPUDevice]] = deviceStr
+		(*annoinput)[util.SupportDevices[NvidiaGPUDevice]] = deviceStr
+		klog.V(5).Infof("pod add notation key [%s], values is [%s]", util.InRequestDevices[NvidiaGPUDevice], deviceStr)
+		klog.V(5).Infof("pod add notation key [%s], values is [%s]", util.SupportDevices[NvidiaGPUDevice], deviceStr)
 	}
 	return *annoinput
 }

pkg/util/types.go

@@ -23,12 +23,10 @@ import (
 const (
 	//ResourceName = "nvidia.com/gpu"
 	//ResourceName = "hami.io/vgpu"
-	AssignedTimeAnnotations          = "hami.io/vgpu-time"
-	AssignedIDsAnnotations           = "hami.io/vgpu-ids-new"
-	AssignedIDsToAllocateAnnotations = "hami.io/devices-to-allocate"
-	AssignedNodeAnnotations          = "hami.io/vgpu-node"
-	BindTimeAnnotations              = "hami.io/bind-time"
-	DeviceBindPhase                  = "hami.io/bind-phase"
+	AssignedTimeAnnotations = "hami.io/vgpu-time"
+	AssignedNodeAnnotations = "hami.io/vgpu-node"
+	BindTimeAnnotations     = "hami.io/bind-time"
+	DeviceBindPhase         = "hami.io/bind-phase"
 
 	DeviceBindAllocating = "allocating"
 	DeviceBindFailed     = "failed"

pkg/util/util.go

@@ -202,6 +202,7 @@ func DecodeContainerDevices(str string) (ContainerDevices, error) {
 }
 
 func DecodePodDevices(checklist map[string]string, annos map[string]string) (PodDevices, error) {
+	klog.V(5).Infof("checklist is [%+v], annos is [%+v]", checklist, annos)
 	if len(annos) == 0 {
 		return PodDevices{}, nil
 	}
@@ -232,7 +233,7 @@ func GetNextDeviceRequest(dtype string, p v1.Pod) (v1.Container, ContainerDevice
 	if err != nil {
 		return v1.Container{}, ContainerDevices{}, err
 	}
-	klog.InfoS("pdevices", pdevices)
+	klog.Infof("pod annotation decode vaule is %+v", pdevices)
 	res := ContainerDevices{}
 
 	pd, ok := pdevices[dtype]
@@ -325,6 +326,7 @@ func PatchPodAnnotations(pod *v1.Pod, annotations map[string]string) error {
 	if err != nil {
 		return err
 	}
+	klog.V(5).Infof("patch pod %s/%s annotation content is %s", pod.Namespace, pod.Name, string(bytes))
 	_, err = client.GetClient().CoreV1().Pods(pod.Namespace).
 		Patch(context.Background(), pod.Name, k8stypes.StrategicMergePatchType, bytes, metav1.PatchOptions{})
 	if err != nil {

pkg/util/util_test.go

@@ -101,3 +101,73 @@ func TestPodDevicesCoding(t *testing.T) {
 		})
 	}
 }
+
+func Test_DecodePodDevices(t *testing.T) {
+	//DecodePodDevices(checklist map[string]string, annos map[string]string) (PodDevices, error)
+	InRequestDevices["NVIDIA"] = "hami.io/vgpu-devices-to-allocate"
+	SupportDevices["NVIDIA"] = "hami.io/vgpu-devices-allocated"
+	tests := []struct {
+		name string
+		args struct {
+			checklist map[string]string
+			annos     map[string]string
+		}
+		want    PodDevices
+		wantErr error
+	}{
+		{
+			name: "annos len is 0",
+			args: struct {
+				checklist map[string]string
+				annos     map[string]string
+			}{
+				checklist: map[string]string{},
+				annos:     make(map[string]string),
+			},
+			want:    PodDevices{},
+			wantErr: nil,
+		},
+		{
+			name: "annos having two device",
+			args: struct {
+				checklist map[string]string
+				annos     map[string]string
+			}{
+				checklist: InRequestDevices,
+				annos: map[string]string{
+					InRequestDevices["NVIDIA"]: "GPU-8dcd427f-483b-b48f-d7e5-75fb19a52b76,NVIDIA,500,3:;GPU-ebe7c3f7-303d-558d-435e-99a160631fe4,NVIDIA,500,3:;",
+					SupportDevices["NVIDIA"]:   "GPU-8dcd427f-483b-b48f-d7e5-75fb19a52b76,NVIDIA,500,3:;GPU-ebe7c3f7-303d-558d-435e-99a160631fe4,NVIDIA,500,3:;",
+				},
+			},
+			want: PodDevices{
+				"NVIDIA": {
+					{
+						{
+							UUID:      "GPU-8dcd427f-483b-b48f-d7e5-75fb19a52b76",
+							Type:      "NVIDIA",
+							Usedmem:   500,
+							Usedcores: 3,
+						},
+					},
+					{
+						{
+							UUID:      "GPU-ebe7c3f7-303d-558d-435e-99a160631fe4",
+							Type:      "NVIDIA",
+							Usedmem:   500,
+							Usedcores: 3,
+						},
+					},
+				},
+			},
+			wantErr: nil,
+		},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			got, gotErr := DecodePodDevices(test.args.checklist, test.args.annos)
+			assert.DeepEqual(t, test.wantErr, gotErr)
+			assert.DeepEqual(t, test.want, got)
+		})
+	}
+}