Scheduler support for hugepages

pkg/api/helper/qos/BUILD

@@ -10,6 +10,7 @@ go_library(
     srcs = ["qos.go"],
     deps = [
         "//pkg/api:go_default_library",
+        "//pkg/api/helper:go_default_library",
         "//vendor/k8s.io/apimachinery/pkg/api/resource:go_default_library",
         "//vendor/k8s.io/apimachinery/pkg/util/sets:go_default_library",
     ],

pkg/api/helper/qos/qos.go

@@ -22,11 +22,16 @@ import (
 	"k8s.io/apimachinery/pkg/api/resource"
 	"k8s.io/apimachinery/pkg/util/sets"
 	"k8s.io/kubernetes/pkg/api"
+	"k8s.io/kubernetes/pkg/api/helper"
 )
 
 // supportedComputeResources is the list of compute resources for with QoS is supported.
 var supportedQoSComputeResources = sets.NewString(string(api.ResourceCPU), string(api.ResourceMemory))
 
+func isSupportedQoSComputeResource(name api.ResourceName) bool {
+	return supportedQoSComputeResources.Has(string(name)) || helper.IsHugePageResourceName(name)
+}
+
 // GetPodQOS returns the QoS class of a pod.
 // A pod is besteffort if none of its containers have specified any requests or limits.
 // A pod is guaranteed only when requests and limits are specified for all the containers and they are equal.
@@ -39,7 +44,7 @@ func GetPodQOS(pod *api.Pod) api.PodQOSClass {
 	for _, container := range pod.Spec.Containers {
 		// process requests
 		for name, quantity := range container.Resources.Requests {
-			if !supportedQoSComputeResources.Has(string(name)) {
+			if !isSupportedQoSComputeResource(name) {
 				continue
 			}
 			if quantity.Cmp(zeroQuantity) == 1 {
@@ -55,7 +60,7 @@ func GetPodQOS(pod *api.Pod) api.PodQOSClass {
 		// process limits
 		qosLimitsFound := sets.NewString()
 		for name, quantity := range container.Resources.Limits {
-			if !supportedQoSComputeResources.Has(string(name)) {
+			if !isSupportedQoSComputeResource(name) {
 				continue
 			}
 			if quantity.Cmp(zeroQuantity) == 1 {

pkg/api/v1/helper/qos/qos_test.go

@@ -132,7 +132,7 @@ func TestGetPodQOS(t *testing.T) {
 		},
 		{
 			pod: newPod("burstable-hugepages", []v1.Container{
-				newContainer("burstable", getResourceList("0", "0"), addResource("hugepages-2Mi", "1Gi", getResourceList("0", "0"))),
+				newContainer("burstable", addResource("hugepages-2Mi", "1Gi", getResourceList("0", "0")), addResource("hugepages-2Mi", "1Gi", getResourceList("0", "0"))),
 			}),
 			expected: v1.PodQOSBurstable,
 		},
@@ -147,7 +147,7 @@ func TestGetPodQOS(t *testing.T) {
 		k8sv1.Convert_v1_Pod_To_api_Pod(testCase.pod, &pod, nil)
 
 		if actual := qos.GetPodQOS(&pod); api.PodQOSClass(testCase.expected) != actual {
-			t.Errorf("[%d]: invalid qos pod %s, expected: %s, actual: %s", id, testCase.pod.Name, testCase.expected, actual)
+			t.Errorf("[%d]: conversion invalid qos pod %s, expected: %s, actual: %s", id, testCase.pod.Name, testCase.expected, actual)
 		}
 	}
 }

plugin/pkg/scheduler/algorithm/predicates/predicates.go

@@ -509,6 +509,12 @@ func GetResourceRequest(pod *v1.Pod) *schedulercache.Resource {
 						result.SetExtended(rName, value)
 					}
 				}
+				if v1helper.IsHugePageResourceName(rName) {
+					value := rQuantity.Value()
+					if value > result.HugePages[rName] {
+						result.SetHugePages(rName, value)
+					}
+				}
 			}
 		}
 	}
@@ -542,7 +548,12 @@ func PodFitsResources(pod *v1.Pod, meta interface{}, nodeInfo *schedulercache.No
 		// We couldn't parse metadata - fallback to computing it.
 		podRequest = GetResourceRequest(pod)
 	}
-	if podRequest.MilliCPU == 0 && podRequest.Memory == 0 && podRequest.NvidiaGPU == 0 && podRequest.EphemeralStorage == 0 && len(podRequest.ExtendedResources) == 0 {
+	if podRequest.MilliCPU == 0 &&
+		podRequest.Memory == 0 &&
+		podRequest.NvidiaGPU == 0 &&
+		podRequest.EphemeralStorage == 0 &&
+		len(podRequest.ExtendedResources) == 0 &&
+		len(podRequest.HugePages) == 0 {
 		return len(predicateFails) == 0, predicateFails, nil
 	}
 
@@ -567,6 +578,12 @@ func PodFitsResources(pod *v1.Pod, meta interface{}, nodeInfo *schedulercache.No
 		}
 	}
 
+	for rName, rQuant := range podRequest.HugePages {
+		if allocatable.HugePages[rName] < rQuant+nodeInfo.RequestedResource().HugePages[rName] {
+			predicateFails = append(predicateFails, NewInsufficientResourceError(rName, podRequest.HugePages[rName], nodeInfo.RequestedResource().HugePages[rName], allocatable.HugePages[rName]))
+		}
+	}
+
 	if glog.V(10) {
 		if len(predicateFails) == 0 {
 			// We explicitly don't do glog.V(10).Infof() to avoid computing all the parameters if this is

plugin/pkg/scheduler/schedulercache/node_info.go

@@ -72,6 +72,7 @@ type Resource struct {
 	// explicitly as int, to avoid conversions and improve performance.
 	AllowedPodNumber  int
 	ExtendedResources map[v1.ResourceName]int64
+	HugePages         map[v1.ResourceName]int64
 }
 
 // New creates a Resource from ResourceList
@@ -103,6 +104,9 @@ func (r *Resource) Add(rl v1.ResourceList) {
 			if v1helper.IsExtendedResourceName(rName) {
 				r.AddExtended(rName, rQuant.Value())
 			}
+			if v1helper.IsHugePageResourceName(rName) {
+				r.AddHugePages(rName, rQuant.Value())
+			}
 		}
 	}
 }
@@ -118,6 +122,9 @@ func (r *Resource) ResourceList() v1.ResourceList {
 	for rName, rQuant := range r.ExtendedResources {
 		result[rName] = *resource.NewQuantity(rQuant, resource.DecimalSI)
 	}
+	for rName, rQuant := range r.HugePages {
+		result[rName] = *resource.NewQuantity(rQuant, resource.BinarySI)
+	}
 	return result
 }
 
@@ -135,6 +142,12 @@ func (r *Resource) Clone() *Resource {
 			res.ExtendedResources[k] = v
 		}
 	}
+	if r.HugePages != nil {
+		res.HugePages = make(map[v1.ResourceName]int64)
+		for k, v := range r.HugePages {
+			res.HugePages[k] = v
+		}
+	}
 	return res
 }
 
@@ -150,6 +163,18 @@ func (r *Resource) SetExtended(name v1.ResourceName, quantity int64) {
 	r.ExtendedResources[name] = quantity
 }
 
+func (r *Resource) AddHugePages(name v1.ResourceName, quantity int64) {
+	r.SetHugePages(name, r.HugePages[name]+quantity)
+}
+
+func (r *Resource) SetHugePages(name v1.ResourceName, quantity int64) {
+	// Lazily allocate hugepages resource map.
+	if r.HugePages == nil {
+		r.HugePages = map[v1.ResourceName]int64{}
+	}
+	r.HugePages[name] = quantity
+}
+
 // NewNodeInfo returns a ready to use empty NodeInfo object.
 // If any pods are given in arguments, their information will be aggregated in
 // the returned object.
@@ -307,6 +332,12 @@ func (n *NodeInfo) addPod(pod *v1.Pod) {
 	for rName, rQuant := range res.ExtendedResources {
 		n.requestedResource.ExtendedResources[rName] += rQuant
 	}
+	if n.requestedResource.HugePages == nil && len(res.HugePages) > 0 {
+		n.requestedResource.HugePages = map[v1.ResourceName]int64{}
+	}
+	for rName, rQuant := range res.HugePages {
+		n.requestedResource.HugePages[rName] += rQuant
+	}
 	n.nonzeroRequest.MilliCPU += non0_cpu
 	n.nonzeroRequest.Memory += non0_mem
 	n.pods = append(n.pods, pod)
@@ -362,6 +393,12 @@ func (n *NodeInfo) removePod(pod *v1.Pod) error {
 			for rName, rQuant := range res.ExtendedResources {
 				n.requestedResource.ExtendedResources[rName] -= rQuant
 			}
+			if len(res.HugePages) > 0 && n.requestedResource.HugePages == nil {
+				n.requestedResource.HugePages = map[v1.ResourceName]int64{}
+			}
+			for rName, rQuant := range res.HugePages {
+				n.requestedResource.HugePages[rName] -= rQuant
+			}
 			n.nonzeroRequest.MilliCPU -= non0_cpu
 			n.nonzeroRequest.Memory -= non0_mem