Align concatenate labels handling with gpu plugin

All labels end with an alpha numeric character and
subsequent labels start with a Z char.

Co-authored-by: Tuomas Katila <[email protected]>
Co-authored-by: Ukri Niemimuukko <[email protected]>

Author: 2 people

gpu-aware-scheduling/pkg/gpuscheduler/scheduler.go

@@ -778,9 +778,9 @@ func createGPUMaps(pod *v1.Pod, node *v1.Node, allGPUs []string) []map[string]bo
 	}
 
 	_, singleNumaRequested := pod.Annotations[singleNumaAnnotationName]
-	gpuNumaInformation, nodeHasNumaInfo := node.Labels[numaMappingLabel]
+	gpuNumaInformation := concatenateSplitLabel(node, numaMappingLabel)
 
-	if singleNumaRequested && nodeHasNumaInfo {
+	if singleNumaRequested && len(gpuNumaInformation) > 0 {
 		numaGroups := strings.Split(gpuNumaInformation, "_")
 
 		for _, numaGroup := range numaGroups {

gpu-aware-scheduling/pkg/gpuscheduler/scheduler_test.go

@@ -688,9 +688,9 @@ func TestGetNodeGPUListFromGpuNumbers(t *testing.T) {
 		node := v1.Node{
 			ObjectMeta: metav1.ObjectMeta{
 				Labels: map[string]string{
-					gpuNumbersLabel:       "0.1.2.",
-					gpuNumbersLabel + "2": "5.8.9.",
-					gpuNumbersLabel + "3": "10"},
+					gpuNumbersLabel:       "0.1.2",
+					gpuNumbersLabel + "2": "Z.5.8.9",
+					gpuNumbersLabel + "3": "Z.10"},
 			},
 		}
 
@@ -897,7 +897,8 @@ func TestFilterWithXeLinkedDisabledTiles(t *testing.T) {
 					Labels: map[string]string{
 						"gpu.intel.com/gpu-numbers": "0.1.2.3",
 						"gpu.intel.com/tiles":       "4",
-						xeLinksLabel:                "0.0-1.0_1.0-0.0_2.1-3.2_3.2-2.1",
+						xeLinksLabel:                "0.0-1.0_1.0-0.0_2.1",
+						xeLinksLabel + "2":          "Z-3.2_3.2-2.1",
 					},
 				},
 				Status: v1.NodeStatus{
@@ -1031,7 +1032,8 @@ func TestRunSchedulingLogicWithMultiContainerXelinkedTileResourceReq(t *testing.
 	testCases := []testCase{
 		{
 			extraLabels: map[string]string{
-				xeLinksLabel: "0.0-1.0_1.0-0.0_2.1-3.2_3.2-2.1",
+				xeLinksLabel:       "0.0-1.0_1.0-0.0",
+				xeLinksLabel + "2": "Z_2.1-3.2_3.2-2.1",
 			},
 			extraAnnotations:       map[string]string{xelinkAnnotationName: trueValueString},
 			description:            "4 card xe-linked success case",
@@ -1042,8 +1044,9 @@ func TestRunSchedulingLogicWithMultiContainerXelinkedTileResourceReq(t *testing.
 		},
 		{
 			extraLabels: map[string]string{
-				xeLinksLabel:     "0.0-1.0_1.0-0.0_2.1-3.2_3.2-2.1",
-				numaMappingLabel: "0-0.1_1-2.3",
+				xeLinksLabel:           "0.0-1.0_1.0-0.0_2.1-3.2_3.2-2.1",
+				numaMappingLabel:       "0-0.1_1",
+				numaMappingLabel + "2": "Z-2.3",
 			},
 			extraAnnotations: map[string]string{
 				xelinkAnnotationName:     trueValueString,

gpu-aware-scheduling/pkg/gpuscheduler/utils.go

@@ -25,6 +25,7 @@ const (
 	desiredIntBits    = 16
 	regexDesiredCount = 3
 	regexXeLinkCount  = 5
+	labelControlChar  = "Z"
 )
 
 // Globals for compiled regexps. No other global types here!
@@ -213,12 +214,20 @@ func isGPUInPCIGroup(gpuName, pciGroupGPUName string, node *v1.Node) bool {
 
 // concatenateSplitLabel returns the given label value and concatenates any
 // additional values for label names with a running number postfix starting with "2".
+// Subsequent values should start with the control character 'Z'.
 func concatenateSplitLabel(node *v1.Node, labelName string) string {
 	postFix := 2
 	value := node.Labels[labelName]
 
 	for continuingLabelValue, ok := node.Labels[labelName+strconv.Itoa(postFix)]; ok; {
-		value += continuingLabelValue
+		if !strings.HasPrefix(continuingLabelValue, labelControlChar) {
+			klog.Warningf("concatenated chuck has invalid prefix: %s", continuingLabelValue[:len(labelControlChar)])
+
+			return ""
+		} else {
+			value += continuingLabelValue[len(labelControlChar):]
+		}
+
 		postFix++
 		continuingLabelValue, ok = node.Labels[labelName+strconv.Itoa(postFix)]
 	}
@@ -348,7 +357,7 @@ func reorderPreferredTilesFirst(tiles []int, preferred []int) []int {
 func getXeLinkedTiles(gpuName string, node *v1.Node) map[int]bool {
 	xeLinkedTiles := map[int]bool{}
 
-	xeLinkLabelValue := node.Labels[xeLinksLabel]
+	xeLinkLabelValue := concatenateSplitLabel(node, xeLinksLabel)
 	lZeroDeviceID := gpuNameToLZeroDeviceID(gpuName, node)
 
 	if lZeroDeviceID == -1 || xeLinkLabelValue == "" {
@@ -410,7 +419,7 @@ func parseXeLink(link string) (linkInfo, error) {
 }
 
 func getXeLinkedGPUInfo(gpuName string, tileIndex int, node *v1.Node) (string, int) {
-	xeLinkLabelValue := node.Labels[xeLinksLabel]
+	xeLinkLabelValue := concatenateSplitLabel(node, xeLinksLabel)
 	lZeroDeviceID := gpuNameToLZeroDeviceID(gpuName, node)
 
 	if lZeroDeviceID == -1 || xeLinkLabelValue == "" {
@@ -456,7 +465,7 @@ func lZeroDeviceIDToGpuName(lZeroID int, node *v1.Node) string {
 }
 
 func numSortedGpuNums(node *v1.Node) []string {
-	gpuNums := node.Labels[gpuNumbersLabel]
+	gpuNums := concatenateSplitLabel(node, gpuNumbersLabel)
 
 	gpuNumSlice := strings.Split(gpuNums, ".")
 

gpu-aware-scheduling/pkg/gpuscheduler/utils_test.go

@@ -130,9 +130,9 @@ func TestPCIGroups(t *testing.T) {
 
 	Convey("When the GPU belongs to a PCI Group with multiple group labels", t, func() {
 		node := getMockNode(1, 1)
-		node.Labels[pciGroupLabel] = "0.1_2.3.4_"
-		node.Labels[pciGroupLabel+"2"] = "5.6_7.8_11.12_"
-		node.Labels[pciGroupLabel+"3"] = "9.10"
+		node.Labels[pciGroupLabel] = "0.1_2.3.4"
+		node.Labels[pciGroupLabel+"2"] = "Z_5.6_7.8_11.12"
+		node.Labels[pciGroupLabel+"3"] = "Z_9.10"
 		So(getPCIGroup(node, "card6"), ShouldResemble, []string{"5", "6"})
 		So(getPCIGroup(node, "card9"), ShouldResemble, []string{"9", "10"})
 		So(getPCIGroup(node, "card20"), ShouldResemble, []string{})
@@ -326,8 +326,8 @@ func TestConcatenateSplitLabel(t *testing.T) {
 	Convey("When the label is split, it can be concatenated", t, func() {
 		node := getMockNode(1, 1)
 		node.Labels[pciGroupLabel] = "foo"
-		node.Labels[pciGroupLabel+"2"] = "bar"
-		node.Labels[pciGroupLabel+"3"] = "ber"
+		node.Labels[pciGroupLabel+"2"] = "Zbar"
+		node.Labels[pciGroupLabel+"3"] = "Zber"
 		result := concatenateSplitLabel(node, pciGroupLabel)
 		So(result, ShouldEqual, "foobarber")
 	})