Topology Manager Policy: single-numa-node

cmd/kubelet/app/options/options.go

@@ -518,7 +518,7 @@ func AddKubeletConfigFlags(mainfs *pflag.FlagSet, c *kubeletconfig.KubeletConfig
 	fs.StringVar(&c.CPUManagerPolicy, "cpu-manager-policy", c.CPUManagerPolicy, "CPU Manager policy to use. Possible values: 'none', 'static'. Default: 'none'")
 	fs.DurationVar(&c.CPUManagerReconcilePeriod.Duration, "cpu-manager-reconcile-period", c.CPUManagerReconcilePeriod.Duration, "<Warning: Alpha feature> CPU Manager reconciliation period. Examples: '10s', or '1m'. If not supplied, defaults to `NodeStatusUpdateFrequency`")
 	fs.Var(cliflag.NewMapStringString(&c.QOSReserved), "qos-reserved", "<Warning: Alpha feature> A set of ResourceName=Percentage (e.g. memory=50%) pairs that describe how pod resource requests are reserved at the QoS level. Currently only memory is supported. Requires the QOSReserved feature gate to be enabled.")
-	fs.StringVar(&c.TopologyManagerPolicy, "topology-manager-policy", c.TopologyManagerPolicy, "Topology Manager policy to use. Possible values: 'none', 'best-effort', 'restricted'.")
+	fs.StringVar(&c.TopologyManagerPolicy, "topology-manager-policy", c.TopologyManagerPolicy, "Topology Manager policy to use. Possible values: 'none', 'best-effort', 'restricted', 'single-numa-node'.")
 	fs.DurationVar(&c.RuntimeRequestTimeout.Duration, "runtime-request-timeout", c.RuntimeRequestTimeout.Duration, "Timeout of all runtime requests except long running request - pull, logs, exec and attach. When timeout exceeded, kubelet will cancel the request, throw out an error and retry later.")
 	fs.StringVar(&c.HairpinMode, "hairpin-mode", c.HairpinMode, "How should the kubelet setup hairpin NAT. This allows endpoints of a Service to loadbalance back to themselves if they should try to access their own Service. Valid values are \"promiscuous-bridge\", \"hairpin-veth\" and \"none\".")
 	fs.Int32Var(&c.MaxPods, "max-pods", c.MaxPods, "Number of Pods that can run on this Kubelet.")

pkg/kubelet/apis/config/types.go

@@ -55,14 +55,17 @@ const (
 	// watches to observe changes to objects that are in its interest.
 	WatchChangeDetectionStrategy ResourceChangeDetectionStrategy = "Watch"
 	// RestrictedTopologyManagerPolicy is a mode in which kubelet only allows
-	// pods with a single NUMA alignment of CPU and device resources.
+	// pods with optimal NUMA node alignment for requested resources
 	RestrictedTopologyManagerPolicy = "restricted"
 	// BestEffortTopologyManagerPolicy is a mode in which kubelet will favour
 	// pods with NUMA alignment of CPU and device resources.
 	BestEffortTopologyManagerPolicy = "best-effort"
 	// NoneTopologyManager Policy is a mode in which kubelet has no knowledge
 	// of NUMA alignment of a pod's CPU and device resources.
 	NoneTopologyManagerPolicy = "none"
+	// SingleNumaNodeTopologyManager Policy iis a mode in which kubelet only allows
+	// pods with a single NUMA alignment of CPU and device resources.
+	SingleNumaNodeTopologyManager = "single-numa-node"
 )
 
 // +k8s:deepcopy-gen:interfaces=k8s.io/apimachinery/pkg/runtime.Object

pkg/kubelet/cm/topologymanager/BUILD

@@ -8,6 +8,7 @@ go_library(
         "policy_best_effort.go",
         "policy_none.go",
         "policy_restricted.go",
+        "policy_single_numa_node.go",
         "topology_manager.go",
     ],
     importpath = "k8s.io/kubernetes/pkg/kubelet/cm/topologymanager",
@@ -45,6 +46,7 @@ go_test(
         "policy_best_effort_test.go",
         "policy_none_test.go",
         "policy_restricted_test.go",
+        "policy_single_numa_node_test.go",
         "topology_manager_test.go",
     ],
     embed = [":go_default_library"],

pkg/kubelet/cm/topologymanager/policy.go

@@ -25,5 +25,5 @@ type Policy interface {
 	//Returns Policy Name
 	Name() string
 	//Returns Pod Admit Handler Response based on hints and policy type
-	CanAdmitPodResult(admit bool) lifecycle.PodAdmitResult
+	CanAdmitPodResult(hint *TopologyHint) lifecycle.PodAdmitResult
 }

pkg/kubelet/cm/topologymanager/policy_best_effort.go

@@ -36,7 +36,7 @@ func (p *bestEffortPolicy) Name() string {
 	return PolicyBestEffort
 }
 
-func (p *bestEffortPolicy) CanAdmitPodResult(admit bool) lifecycle.PodAdmitResult {
+func (p *bestEffortPolicy) CanAdmitPodResult(hint *TopologyHint) lifecycle.PodAdmitResult {
 	return lifecycle.PodAdmitResult{
 		Admit: true,
 	}

pkg/kubelet/cm/topologymanager/policy_best_effort_test.go

@@ -23,25 +23,24 @@ import (
 func TestPolicyBestEffortCanAdmitPodResult(t *testing.T) {
 	tcases := []struct {
 		name     string
-		admit    bool
+		hint     TopologyHint
 		expected bool
 	}{
 		{
-			name:     "Affinity is set to false in topology hints",
-			admit:    false,
+			name:     "Preferred is set to false in topology hints",
+			hint:     TopologyHint{nil, false},
 			expected: true,
 		},
 		{
-			name:     "Affinity is set to true in topology hints",
-			admit:    true,
+			name:     "Preferred is set to true in topology hints",
+			hint:     TopologyHint{nil, true},
 			expected: true,
 		},
 	}
 
 	for _, tc := range tcases {
 		policy := NewBestEffortPolicy()
-		admit := tc.admit
-		result := policy.CanAdmitPodResult(admit)
+		result := policy.CanAdmitPodResult(&tc.hint)
 
 		if result.Admit != tc.expected {
 			t.Errorf("Expected Admit field in result to be %t, got %t", tc.expected, result.Admit)

pkg/kubelet/cm/topologymanager/policy_none.go

@@ -36,7 +36,7 @@ func (p *nonePolicy) Name() string {
 	return PolicyNone
 }
 
-func (p *nonePolicy) CanAdmitPodResult(admit bool) lifecycle.PodAdmitResult {
+func (p *nonePolicy) CanAdmitPodResult(hint *TopologyHint) lifecycle.PodAdmitResult {
 	return lifecycle.PodAdmitResult{
 		Admit: true,
 	}

pkg/kubelet/cm/topologymanager/policy_none_test.go

@@ -41,25 +41,24 @@ func TestName(t *testing.T) {
 func TestPolicyNoneCanAdmitPodResult(t *testing.T) {
 	tcases := []struct {
 		name     string
-		admit    bool
+		hint     TopologyHint
 		expected bool
 	}{
 		{
-			name:     "Affinity is set to false in topology hints",
-			admit:    false,
+			name:     "Preferred is set to false in topology hints",
+			hint:     TopologyHint{nil, false},
 			expected: true,
 		},
 		{
-			name:     "Affinity is set to true in topology hints",
-			admit:    true,
+			name:     "Preferred is set to true in topology hints",
+			hint:     TopologyHint{nil, true},
 			expected: true,
 		},
 	}
 
 	for _, tc := range tcases {
 		policy := NewNonePolicy()
-		admit := tc.admit
-		result := policy.CanAdmitPodResult(admit)
+		result := policy.CanAdmitPodResult(&tc.hint)
 
 		if result.Admit != tc.expected {
 			t.Errorf("Expected Admit field in result to be %t, got %t", tc.expected, result.Admit)

pkg/kubelet/cm/topologymanager/policy_restricted.go

@@ -36,8 +36,8 @@ func (p *restrictedPolicy) Name() string {
 	return PolicyRestricted
 }
 
-func (p *restrictedPolicy) CanAdmitPodResult(admit bool) lifecycle.PodAdmitResult {
-	if !admit {
+func (p *restrictedPolicy) CanAdmitPodResult(hint *TopologyHint) lifecycle.PodAdmitResult {
+	if !hint.Preferred {
 		return lifecycle.PodAdmitResult{
 			Admit:   false,
 			Reason:  "Topology Affinity Error",

pkg/kubelet/cm/topologymanager/policy_restricted_test.go

@@ -23,25 +23,24 @@ import (
 func TestPolicyRestrictedCanAdmitPodResult(t *testing.T) {
 	tcases := []struct {
 		name     string
-		admit    bool
+		hint     TopologyHint
 		expected bool
 	}{
 		{
-			name:     "Affinity is set to false in topology hints",
-			admit:    false,
+			name:     "Preferred is set to false in topology hints",
+			hint:     TopologyHint{nil, false},
 			expected: false,
 		},
 		{
-			name:     "Affinity is set to true in topology hints",
-			admit:    true,
+			name:     "Preferred is set to true in topology hints",
+			hint:     TopologyHint{nil, true},
 			expected: true,
 		},
 	}
 
 	for _, tc := range tcases {
 		policy := NewRestrictedPolicy()
-		admit := tc.admit
-		result := policy.CanAdmitPodResult(admit)
+		result := policy.CanAdmitPodResult(&tc.hint)
 
 		if result.Admit != tc.expected {
 			t.Errorf("Expected Admit field in result to be %t, got %t", tc.expected, result.Admit)

pkg/kubelet/cm/topologymanager/policy_single_numa_node.go

@@ -0,0 +1,50 @@
+/*
+Copyright 2019 The Kubernetes Authors.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+package topologymanager
+
+import (
+	"k8s.io/kubernetes/pkg/kubelet/lifecycle"
+)
+
+type singleNumaNodePolicy struct{}
+
+var _ Policy = &singleNumaNodePolicy{}
+
+// PolicySingleNumaNode policy name.
+const PolicySingleNumaNode string = "single-numa-node"
+
+// NewSingleNumaNodePolicy returns single-numa-node policy.
+func NewSingleNumaNodePolicy() Policy {
+	return &singleNumaNodePolicy{}
+}
+
+func (p *singleNumaNodePolicy) Name() string {
+	return PolicySingleNumaNode
+}
+
+func (p *singleNumaNodePolicy) CanAdmitPodResult(hint *TopologyHint) lifecycle.PodAdmitResult {
+	if !hint.Preferred || hint.NUMANodeAffinity.Count() > 1 {
+		return lifecycle.PodAdmitResult{
+			Admit:   false,
+			Reason:  "Topology Affinity Error",
+			Message: "Resources cannot be allocated with Topology Locality",
+		}
+	}
+	return lifecycle.PodAdmitResult{
+		Admit: true,
+	}
+}

pkg/kubelet/cm/topologymanager/policy_single_numa_node_test.go

@@ -0,0 +1,63 @@
+/*
+Copyright 2019 The Kubernetes Authors.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+package topologymanager
+
+import (
+	"testing"
+)
+
+func TestPolicySingleNumaNodeCanAdmitPodResult(t *testing.T) {
+	tcases := []struct {
+		name     string
+		hint     TopologyHint
+		expected bool
+	}{
+		{
+			name:     "Preferred is set to false in topology hints",
+			hint:     TopologyHint{nil, false},
+			expected: false,
+		},
+		{
+			name:     "NUMANodeAffinity has multiple NUMA Nodes masked in topology hints",
+			hint:     TopologyHint{NewTestSocketMask(0, 1), true},
+			expected: false,
+		},
+		{
+			name:     "NUMANodeAffinity has one NUMA Node masked in topology hints",
+			hint:     TopologyHint{NewTestSocketMask(0), true},
+			expected: true,
+		},
+	}
+
+	for _, tc := range tcases {
+		policy := NewSingleNumaNodePolicy()
+		result := policy.CanAdmitPodResult(&tc.hint)
+
+		if result.Admit != tc.expected {
+			t.Errorf("Expected Admit field in result to be %t, got %t", tc.expected, result.Admit)
+		}
+
+		if tc.expected == false {
+			if len(result.Reason) == 0 {
+				t.Errorf("Expected Reason field to be not empty")
+			}
+			if len(result.Message) == 0 {
+				t.Errorf("Expected Message field to be not empty")
+			}
+		}
+	}
+}

pkg/kubelet/cm/topologymanager/topology_manager.go

@@ -103,6 +103,9 @@ func NewManager(numaNodeInfo cputopology.NUMANodeInfo, topologyPolicyName string
 	case PolicyRestricted:
 		policy = NewRestrictedPolicy()
 
+	case PolicySingleNumaNode:
+		policy = NewSingleNumaNodePolicy()
+
 	default:
 		return nil, fmt.Errorf("unknown policy: \"%s\"", topologyPolicyName)
 	}
@@ -228,6 +231,11 @@ func (m *manager) calculateAffinity(pod v1.Pod, container v1.Container) Topology
 				if !hint.Preferred {
 					preferred = false
 				}
+				// Special case PolicySingleNumaNode to only prefer hints where
+				// all providers have a single NUMA affinity set.
+				if m.policy != nil && m.policy.Name() == PolicySingleNumaNode && hint.NUMANodeAffinity.Count() > 1 {
+					preferred = false
+				}
 				numaAffinities = append(numaAffinities, hint.NUMANodeAffinity)
 			}
 		}
@@ -308,7 +316,7 @@ func (m *manager) Admit(attrs *lifecycle.PodAdmitAttributes) lifecycle.PodAdmitR
 	if pod.Status.QOSClass == v1.PodQOSGuaranteed {
 		for _, container := range append(pod.Spec.InitContainers, pod.Spec.Containers...) {
 			result := m.calculateAffinity(*pod, container)
-			admitPod := m.policy.CanAdmitPodResult(result.Preferred)
+			admitPod := m.policy.CanAdmitPodResult(&result)
 			if !admitPod.Admit {
 				return admitPod
 			}

pkg/kubelet/cm/topologymanager/topology_manager_test.go

@@ -111,6 +111,7 @@ func TestCalculateAffinity(t *testing.T) {
 		name     string
 		hp       []HintProvider
 		expected TopologyHint
+		policy   Policy
 	}{
 		{
 			name: "TopologyHint not set",
@@ -655,10 +656,45 @@ func TestCalculateAffinity(t *testing.T) {
 				Preferred:        true,
 			},
 		},
+		{
+			name:   "Special cased PolicySingleNumaNode for single NUMA hint generation",
+			policy: NewSingleNumaNodePolicy(),
+			hp: []HintProvider{
+				&mockHintProvider{
+					map[string][]TopologyHint{
+						"resource1": {
+							{
+								NUMANodeAffinity: NewTestSocketMask(0, 1),
+								Preferred:        true,
+							},
+						},
+						"resource2": {
+							{
+								NUMANodeAffinity: NewTestSocketMask(0),
+								Preferred:        true,
+							},
+							{
+								NUMANodeAffinity: NewTestSocketMask(1),
+								Preferred:        true,
+							},
+							{
+								NUMANodeAffinity: NewTestSocketMask(0, 1),
+								Preferred:        false,
+							},
+						},
+					},
+				},
+			},
+			expected: TopologyHint{
+				NUMANodeAffinity: NewTestSocketMask(0),
+				Preferred:        false,
+			},
+		},
 	}
 
 	for _, tc := range tcases {
 		mngr := manager{
+			policy:        tc.policy,
 			hintProviders: tc.hp,
 			numaNodes:     numaNodes,
 		}

staging/src/k8s.io/kubelet/config/v1beta1/types.go

@@ -55,14 +55,17 @@ const (
 	// watches to observe changes to objects that are in its interest.
 	WatchChangeDetectionStrategy ResourceChangeDetectionStrategy = "Watch"
 	// RestrictedTopologyManagerPolicy is a mode in which kubelet only allows
-	// pods with a single NUMA alignment of CPU and device resources.
+	// pods with optimal NUMA node alignment for requested resources
 	RestrictedTopologyManagerPolicy = "restricted"
 	// BestEffortTopologyManagerPolicy is a mode in which kubelet will favour
 	// pods with NUMA alignment of CPU and device resources.
 	BestEffortTopologyManagerPolicy = "best-effort"
 	// NoneTopologyManager Policy is a mode in which kubelet has no knowledge
 	// of NUMA alignment of a pod's CPU and device resources.
 	NoneTopologyManagerPolicy = "none"
+	// SingleNumaNodeTopologyManager Policy iis a mode in which kubelet only allows
+	// pods with a single NUMA alignment of CPU and device resources.
+	SingleNumaNodeTopologyManager = "single-numa-node"
 )
 
 // +k8s:deepcopy-gen:interfaces=k8s.io/apimachinery/pkg/runtime.Object