Merge remote-tracking branch 'origin/master' into self-monitoring

examples/nvidia/use_as_normal.yaml

@@ -0,0 +1,26 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  name: gpu-pod1
+spec:
+  containers:
+    - name: ubuntu-container
+      image: ubuntu:18.04
+      command: ["bash", "-c", "sleep 86400"]
+      resources:
+        limits:
+          nvidia.com/gpu: 2 # requesting 2 vGPUs
+---
+apiVersion: v1
+kind: Pod
+metadata:
+  name: gpu-pod2
+spec:
+  containers:
+    - name: ubuntu-container
+      image: ubuntu:18.04
+      command: ["bash", "-c", "sleep 86400"]
+      resources:
+        limits:
+          nvidia.com/gpu: 2 # requesting 2 vGPUs
+# gpu-pod1 and gpu-pod2 will NOT share the same GPU

examples/nvidia/use_exclusive_card.yaml

@@ -11,4 +11,4 @@ spec:
         limits:
           nvidia.com/gpu: 2 # requesting 2 vGPUs
           nvidia.com/gpumem-percentage: 100 # Each vGPU contains 100% of the entire GPU device memory （Optional,Integer）
-          nvidia.com/gpucores: 100 # Each vGPU uses 30% of the entire GPU cores（Optional,Integer)
+          nvidia.com/gpucores: 100 # Each vGPU uses 100% of the entire GPU cores（Optional,Integer)

pkg/scheduler/policy/gpu_policy.go

@@ -67,9 +67,11 @@ func (ds *DeviceListsScore) ComputeScore(requests util.ContainerDeviceRequests)
 		}
 		mem += container.Memreq
 	}
-	useScore := float32(request+ds.Device.Used) / float32(ds.Device.Count)
+	klog.V(2).Infof("device %s user %d, userCore %d, userMem %d,", ds.Device.ID, ds.Device.Used, ds.Device.Usedcores, ds.Device.Usedmem)
+
+	usedScore := float32(request+ds.Device.Used) / float32(ds.Device.Count)
 	coreScore := float32(core+ds.Device.Usedcores) / float32(ds.Device.Totalcore)
 	memScore := float32(mem+ds.Device.Usedmem) / float32(ds.Device.Totalmem)
-	ds.Score = float32(Weight) * (useScore + coreScore + memScore)
+	ds.Score = float32(Weight) * (usedScore + coreScore + memScore)
 	klog.V(2).Infof("device %s computer score is %f", ds.Device.ID, ds.Score)
 }

pkg/scheduler/policy/node_policy.go

@@ -51,24 +51,24 @@ func (l NodeScoreList) Less(i, j int) bool {
 }
 
 func (ns *NodeScore) ComputeScore(devices DeviceUsageList) {
-	request, core, mem := int32(0), int32(0), int32(0)
 	// current user having request resource
-	user, userCore, userMem := int32(0), int32(0), int32(0)
+	used, usedCore, usedMem := int32(0), int32(0), int32(0)
 	for _, device := range devices.DeviceLists {
-		user += device.Device.Used
-		userCore += device.Device.Usedcores
-		userMem += device.Device.Usedmem
+		used += device.Device.Used
+		usedCore += device.Device.Usedcores
+		usedMem += device.Device.Usedmem
 	}
+	klog.V(2).Infof("node %s used %d, usedCore %d, usedMem %d,", ns.NodeID, used, usedCore, usedMem)
 
 	total, totalCore, totalMem := int32(0), int32(0), int32(0)
 	for _, deviceLists := range devices.DeviceLists {
 		total += deviceLists.Device.Count
 		totalCore += deviceLists.Device.Totalcore
 		totalMem += deviceLists.Device.Totalmem
 	}
-	useScore := float32(request+user) / float32(total)
-	coreScore := float32(core+userCore) / float32(totalCore)
-	memScore := float32(mem+userMem) / float32(totalMem)
+	useScore := float32(used) / float32(total)
+	coreScore := float32(usedCore) / float32(totalCore)
+	memScore := float32(usedMem) / float32(totalMem)
 	ns.Score = float32(Weight) * (useScore + coreScore + memScore)
 	klog.V(2).Infof("node %s computer score is %f", ns.NodeID, ns.Score)
 }

pkg/scheduler/scheduler.go

@@ -388,7 +388,7 @@ func (s *Scheduler) Bind(args extenderv1.ExtenderBindingArgs) (*extenderv1.Exten
 	for _, val := range device.GetDevices() {
 		err = val.LockNode(node, current)
 		if err != nil {
-			goto RelaseNodeLocks
+			goto ReleaseNodeLocks
 		}
 	}
 	/*
@@ -419,7 +419,7 @@ func (s *Scheduler) Bind(args extenderv1.ExtenderBindingArgs) (*extenderv1.Exten
 		klog.Infoln("After Binding Process")
 		return res, nil
 	}
-RelaseNodeLocks:
+ReleaseNodeLocks:
 	klog.InfoS("bind failed", "err", err.Error())
 	for _, val := range device.GetDevices() {
 		val.ReleaseNodeLock(node, current)

pkg/scheduler/score.go

@@ -197,6 +197,7 @@ func (s *Scheduler) calcScore(nodes *map[string]*NodeUsage, nums util.PodDeviceR
 	for nodeID, node := range *nodes {
 		viewStatus(*node)
 		score := policy.NodeScore{NodeID: nodeID, Devices: make(util.PodDevices), Score: 0}
+		score.ComputeScore(node.Devices)
 
 		//This loop is for different container request
 		ctrfit := false
@@ -225,7 +226,6 @@ func (s *Scheduler) calcScore(nodes *map[string]*NodeUsage, nums util.PodDeviceR
 		}
 
 		if ctrfit {
-			score.ComputeScore(node.Devices)
 			res.NodeList = append(res.NodeList, &score)
 		}
 	}

pkg/scheduler/score_test.go

@@ -147,7 +147,7 @@ func Test_calcScore(t *testing.T) {
 									},
 								},
 							},
-							Score: 5.25,
+							Score: 0,
 						},
 					},
 				},
@@ -242,7 +242,7 @@ func Test_calcScore(t *testing.T) {
 									},
 								},
 							},
-							Score: 20.25,
+							Score: 15,
 						},
 					},
 				},
@@ -353,7 +353,7 @@ func Test_calcScore(t *testing.T) {
 									},
 								},
 							},
-							Score: 2.625,
+							Score: 0,
 						},
 					},
 				},
@@ -464,7 +464,7 @@ func Test_calcScore(t *testing.T) {
 									},
 								},
 							},
-							Score: 10.125,
+							Score: 7.5,
 						},
 					},
 				},
@@ -582,7 +582,7 @@ func Test_calcScore(t *testing.T) {
 									},
 								},
 							},
-							Score: 5.25,
+							Score: 0,
 						},
 					},
 				},
@@ -700,7 +700,7 @@ func Test_calcScore(t *testing.T) {
 									},
 								},
 							},
-							Score: 20.25,
+							Score: 15,
 						},
 					},
 				},
@@ -818,7 +818,7 @@ func Test_calcScore(t *testing.T) {
 									},
 								},
 							},
-							Score: 5.25,
+							Score: 0,
 						},
 						{
 							NodeID: "node2",
@@ -835,7 +835,7 @@ func Test_calcScore(t *testing.T) {
 									},
 								},
 							},
-							Score: 5.25,
+							Score: 0,
 						},
 					},
 				},
@@ -953,7 +953,7 @@ func Test_calcScore(t *testing.T) {
 									},
 								},
 							},
-							Score: 20.25,
+							Score: 15,
 						},
 						{
 							NodeID: "node2",
@@ -970,7 +970,7 @@ func Test_calcScore(t *testing.T) {
 									},
 								},
 							},
-							Score: 5.25,
+							Score: 0,
 						},
 					},
 				},
@@ -1081,7 +1081,7 @@ func Test_calcScore(t *testing.T) {
 									},
 								},
 							},
-							Score: 5.25,
+							Score: 0,
 						},
 					},
 				},