Allow admin user to explicitly unschedule the node

docs/node.md

@@ -38,13 +38,16 @@ must have appropriate conditions, see below.
 
 ### Node Condition
 Node Condition describes the conditions of `Running` nodes. Current valid
-conditions are `NodeReachable` and `NodeReady`. In the future, we plan to
-add more. `NodeReachable` means the node can be reached within the cluster.
-`NodeReady` means the kubelet returns StatusOK for HTTP health check. Different
-condition provides different level of understanding for node health. Kubernetes
-will make a comprehensive scheduling decision based on the information. Node
-condition is represented as a json object. For example, the following conditions
-mean the node is reachable from its cluster, but not ready to accept pods:
+conditions are `NodeReachable`, `NodeReady` and `NodeSchedulable`. In the
+future, we plan to add more. `NodeReachable` means the node can be reached
+within the cluster. `NodeReady` means the kubelet returns StatusOK for HTTP
+health check. `NodeSchedulable` means node is allowed to schedule any new
+pods and is controlled by 'unschedulable' field in node spec.
+Different condition provides different level of understanding for node
+health. Kubernetes will make a comprehensive scheduling decision based on the
+information. Node condition is represented as a json object. For example, the
+following conditions mean the node is reachable from its cluster, node is in
+sane state but not allowed to accept new pods:
 ```json
 "conditions": [
   {
@@ -53,8 +56,12 @@ mean the node is reachable from its cluster, but not ready to accept pods:
   },
   {
     "kind": "Ready",
+    "status": "Full",
+  },
+  {
+    "kind": "Schedulable",
     "status": "None",
-  }
+  },
 ]
 ```
 
@@ -125,3 +132,9 @@ A Kubernetes administrator typically uses `kubectl` to manage `Node`. Similar
 to Node Controller, `kubectl` command only creates/deletes node representation.
 Note if Kubernetes is running on cloud provider, `kubectl create` a node will
 be refused if Node Controller has already synchronized nodes from cloud provider.
+Admin can choose to make the node unschedulable using `kubectl`. Unscheduling the node
+will not affect any existing pods on the node but it will disable creation of
+any new pods on the node. Node unschedulable example:
+```
+kubectl update nodes 10.1.2.3 --patch='{"apiVersion": "v1beta1", "unschedulable": true}'
+```

pkg/api/types.go

@@ -782,13 +782,18 @@ type EndpointsList struct {
 
 // NodeSpec describes the attributes that a node is created with.
 type NodeSpec struct {
-	// Capacity represents the available resources of a node
+	// Capacity represents the available resources of a node.
 	Capacity ResourceList `json:"capacity,omitempty"`
+
 	// PodCIDR represents the pod IP range assigned to the node
 	// Note: assigning IP ranges to nodes might need to be revisited when we support migratable IPs.
 	PodCIDR string `json:"podCIDR,omitempty"`
+
 	// External ID of the node assigned by some machine database (e.g. a cloud provider)
 	ExternalID string `json:"externalID,omitempty"`
+
+	// Unschedulable controls node schedulability of new pods. By default node is schedulable.
+	Unschedulable bool `json:"unschedulable,omitempty"`
 }
 
 // NodeSystemInfo is a set of ids/uuids to uniquely identify the node.
@@ -842,6 +847,8 @@ const (
 	NodeReachable NodeConditionType = "Reachable"
 	// NodeReady means the node returns StatusOK for HTTP health check.
 	NodeReady NodeConditionType = "Ready"
+	// NodeSchedulable means the node is ready to accept new pods.
+	NodeSchedulable NodeConditionType = "Schedulable"
 )
 
 type NodeCondition struct {

pkg/api/v1beta1/conversion.go

@@ -711,6 +711,7 @@ func init() {
 			}
 			out.PodCIDR = in.Spec.PodCIDR
 			out.ExternalID = in.Spec.ExternalID
+			out.Unschedulable = in.Spec.Unschedulable
 			return s.Convert(&in.Spec.Capacity, &out.NodeResources.Capacity, 0)
 		},
 		func(in *Minion, out *newer.Node, s conversion.Scope) error {
@@ -742,6 +743,7 @@ func init() {
 			}
 			out.Spec.PodCIDR = in.PodCIDR
 			out.Spec.ExternalID = in.ExternalID
+			out.Spec.Unschedulable = in.Unschedulable
 			return s.Convert(&in.NodeResources.Capacity, &out.Spec.Capacity, 0)
 		},
 

pkg/api/v1beta1/types.go

@@ -677,6 +677,8 @@ const (
 	NodeReachable NodeConditionKind = "Reachable"
 	// NodeReady means the node returns StatusOK for HTTP health check.
 	NodeReady NodeConditionKind = "Ready"
+	// NodeSchedulable means the node is ready to accept new pods.
+	NodeSchedulable NodeConditionKind = "Schedulable"
 )
 
 type NodeCondition struct {
@@ -731,6 +733,8 @@ type Minion struct {
 	NodeResources NodeResources `json:"resources,omitempty" description:"characterization of node resources"`
 	// Pod IP range assigned to the node
 	PodCIDR string `json:"podCIDR,omitempty" description:"IP range assigned to the node"`
+	// Unschedulable controls node schedulability of new pods. By default node is schedulable.
+	Unschedulable bool `json:"unschedulable,omitempty" description:"disable pod scheduling on the node"`
 	// Status describes the current status of a node
 	Status NodeStatus `json:"status,omitempty" description:"current status of node"`
 	// Labels for the node

pkg/api/v1beta2/conversion.go

@@ -631,6 +631,7 @@ func init() {
 			}
 			out.PodCIDR = in.Spec.PodCIDR
 			out.ExternalID = in.Spec.ExternalID
+			out.Unschedulable = in.Spec.Unschedulable
 			return s.Convert(&in.Spec.Capacity, &out.NodeResources.Capacity, 0)
 		},
 		func(in *Minion, out *newer.Node, s conversion.Scope) error {
@@ -662,6 +663,7 @@ func init() {
 			}
 			out.Spec.PodCIDR = in.PodCIDR
 			out.Spec.ExternalID = in.ExternalID
+			out.Spec.Unschedulable = in.Unschedulable
 			return s.Convert(&in.NodeResources.Capacity, &out.Spec.Capacity, 0)
 		},
 

pkg/api/v1beta2/types.go

@@ -690,6 +690,8 @@ const (
 	NodeReachable NodeConditionKind = "Reachable"
 	// NodeReady means the node returns StatusOK for HTTP health check.
 	NodeReady NodeConditionKind = "Ready"
+	// NodeSchedulable means the node is ready to accept new pods.
+	NodeSchedulable NodeConditionKind = "Schedulable"
 )
 
 // Described the conditions of a running node.
@@ -750,6 +752,8 @@ type Minion struct {
 	NodeResources NodeResources `json:"resources,omitempty" description:"characterization of node resources"`
 	// Pod IP range assigned to the node
 	PodCIDR string `json:"podCIDR,omitempty" description:"IP range assigned to the node"`
+	// Unschedulable controls node schedulability of new pods. By default node is schedulable.
+	Unschedulable bool `json:"unschedulable,omitempty" description:"disable pod scheduling on the node"`
 	// Status describes the current status of a node
 	Status NodeStatus `json:"status,omitempty" description:"current status of node"`
 	// Labels for the node

pkg/api/v1beta3/types.go

@@ -813,13 +813,15 @@ type EndpointsList struct {
 
 // NodeSpec describes the attributes that a node is created with.
 type NodeSpec struct {
-	// Capacity represents the available resources of a node
+	// Capacity represents the available resources of a node.
 	// see https://github.com/GoogleCloudPlatform/kubernetes/blob/master/docs/resources.md for more details.
 	Capacity ResourceList `json:"capacity,omitempty" description:"compute resource capacity of the node; https://github.com/GoogleCloudPlatform/kubernetes/blob/master/docs/resources.md"`
 	// PodCIDR represents the pod IP range assigned to the node
 	PodCIDR string `json:"podCIDR,omitempty" description:"pod IP range assigned to the node"`
 	// External ID of the node assigned by some machine database (e.g. a cloud provider)
 	ExternalID string `json:"externalID,omitempty" description:"external ID assigned to the node by some machine database (e.g. a cloud provider)"`
+	// Unschedulable controls node schedulability of new pods. By default node is schedulable.
+	Unschedulable bool `json:"unschedulable,omitempty" description:"disable pod scheduling on the node"`
 }
 
 // NodeSystemInfo is a set of ids/uuids to uniquely identify the node.
@@ -873,6 +875,8 @@ const (
 	NodeReachable NodeConditionType = "Reachable"
 	// NodeReady means the node returns StatusOK for HTTP health check.
 	NodeReady NodeConditionType = "Ready"
+	// NodeSchedulable means the node is ready to accept new pods.
+	NodeSchedulable NodeConditionType = "Schedulable"
 )
 
 type NodeCondition struct {

pkg/api/validation/validation.go

@@ -862,6 +862,8 @@ func ValidateMinionUpdate(oldMinion *api.Node, minion *api.Node) errs.Validation
 	oldMinion.ObjectMeta = minion.ObjectMeta
 	// Allow users to update capacity
 	oldMinion.Spec.Capacity = minion.Spec.Capacity
+	// Allow users to unschedule node
+	oldMinion.Spec.Unschedulable = minion.Spec.Unschedulable
 	// Clear status
 	oldMinion.Status = minion.Status
 

pkg/api/validation/validation_test.go

@@ -2106,6 +2106,21 @@ func TestValidateMinionUpdate(t *testing.T) {
 				Labels: map[string]string{"Foo": "baz"},
 			},
 		}, true},
+		{api.Node{
+			ObjectMeta: api.ObjectMeta{
+				Name: "foo",
+			},
+			Spec: api.NodeSpec{
+				Unschedulable: false,
+			},
+		}, api.Node{
+			ObjectMeta: api.ObjectMeta{
+				Name: "foo",
+			},
+			Spec: api.NodeSpec{
+				Unschedulable: true,
+			},
+		}, true},
 	}
 	for i, test := range tests {
 		errs := ValidateMinionUpdate(&test.oldMinion, &test.minion)

pkg/client/client_test.go

@@ -749,6 +749,7 @@ func TestCreateMinion(t *testing.T) {
 				api.ResourceCPU:    resource.MustParse("1000m"),
 				api.ResourceMemory: resource.MustParse("1Mi"),
 			},
+			Unschedulable: false,
 		},
 	}
 	c := &testClient{
@@ -782,6 +783,7 @@ func TestUpdateMinion(t *testing.T) {
 				api.ResourceCPU:    resource.MustParse("1000m"),
 				api.ResourceMemory: resource.MustParse("1Mi"),
 			},
+			Unschedulable: true,
 		},
 	}
 	c := &testClient{

pkg/cloudprovider/controller/nodecontroller.go

@@ -301,7 +301,7 @@ func (s *NodeController) PopulateAddresses(nodes *api.NodeList) (*api.NodeList,
 	return nodes, nil
 }
 
-// UpdateNodesStatus performs health checking for given list of nodes.
+// UpdateNodesStatus performs various condition checks for given list of nodes.
 func (s *NodeController) UpdateNodesStatus(nodes *api.NodeList) *api.NodeList {
 	var wg sync.WaitGroup
 	wg.Add(len(nodes.Items))
@@ -330,21 +330,14 @@ func (s *NodeController) updateNodeInfo(node *api.Node) error {
 	return nil
 }
 
-// DoCheck performs health checking for given node.
+// DoCheck performs various condition checks for given node.
 func (s *NodeController) DoCheck(node *api.Node) []api.NodeCondition {
 	var conditions []api.NodeCondition
 
 	// Check Condition: NodeReady. TODO: More node conditions.
 	oldReadyCondition := s.getCondition(node, api.NodeReady)
 	newReadyCondition := s.checkNodeReady(node)
-	if oldReadyCondition != nil && oldReadyCondition.Status == newReadyCondition.Status {
-		// If node status doesn't change, transition time is same as last time.
-		newReadyCondition.LastTransitionTime = oldReadyCondition.LastTransitionTime
-	} else {
-		// Set transition time to Now() if node status changes or `oldReadyCondition` is nil, which
-		// happens only when the node is checked for the first time.
-		newReadyCondition.LastTransitionTime = util.Now()
-	}
+	s.updateLastTransitionTime(oldReadyCondition, newReadyCondition)
 
 	if newReadyCondition.Status != api.ConditionFull {
 		// Node is not ready for this probe, we need to check if pods need to be deleted.
@@ -355,12 +348,48 @@ func (s *NodeController) DoCheck(node *api.Node) []api.NodeCondition {
 			s.deletePods(node.Name)
 		}
 	}
-
 	conditions = append(conditions, *newReadyCondition)
 
+	// Check Condition: NodeSchedulable
+	oldSchedulableCondition := s.getCondition(node, api.NodeSchedulable)
+	newSchedulableCondition := s.checkNodeSchedulable(node)
+	s.updateLastTransitionTime(oldSchedulableCondition, newSchedulableCondition)
+	conditions = append(conditions, *newSchedulableCondition)
+
 	return conditions
 }
 
+// updateLastTransitionTime updates LastTransitionTime for the newCondition based on oldCondition.
+func (s *NodeController) updateLastTransitionTime(oldCondition, newCondition *api.NodeCondition) {
+	if oldCondition != nil && oldCondition.Status == newCondition.Status {
+		// If node status doesn't change, transition time is same as last time.
+		newCondition.LastTransitionTime = oldCondition.LastTransitionTime
+	} else {
+		// Set transition time to Now() if node status changes or `oldCondition` is nil, which
+		// happens only when the node is checked for the first time.
+		newCondition.LastTransitionTime = util.Now()
+	}
+}
+
+// checkNodeSchedulable checks node schedulable condition, without transition timestamp set.
+func (s *NodeController) checkNodeSchedulable(node *api.Node) *api.NodeCondition {
+	if node.Spec.Unschedulable {
+		return &api.NodeCondition{
+			Type:          api.NodeSchedulable,
+			Status:        api.ConditionNone,
+			Reason:        "User marked unschedulable during node create/update",
+			LastProbeTime: util.Now(),
+		}
+	} else {
+		return &api.NodeCondition{
+			Type:          api.NodeSchedulable,
+			Status:        api.ConditionFull,
+			Reason:        "Node is schedulable by default",
+			LastProbeTime: util.Now(),
+		}
+	}
+}
+
 // checkNodeReady checks raw node ready condition, without transition timestamp set.
 func (s *NodeController) checkNodeReady(node *api.Node) *api.NodeCondition {
 	switch status, err := s.kubeletClient.HealthCheck(node.Name); {

pkg/cloudprovider/controller/nodecontroller_test.go

@@ -550,13 +550,15 @@ func TestSyncCloudDeletePods(t *testing.T) {
 	}
 }
 
-func TestHealthCheckNode(t *testing.T) {
+func TestNodeConditionsCheck(t *testing.T) {
 	table := []struct {
 		node               *api.Node
 		fakeKubeletClient  *FakeKubeletClient
 		expectedConditions []api.NodeCondition
 	}{
 		{
+			// Node with default spec and kubelet /healthz probe returns success.
+			// Expected node condition to be ready and marked schedulable.
 			node: newNode("node0"),
 			fakeKubeletClient: &FakeKubeletClient{
 				Status: probe.Success,
@@ -568,10 +570,17 @@ func TestHealthCheckNode(t *testing.T) {
 					Status: api.ConditionFull,
 					Reason: "Node health check succeeded: kubelet /healthz endpoint returns ok",
 				},
+				{
+					Type:   api.NodeSchedulable,
+					Status: api.ConditionFull,
+					Reason: "Node is schedulable by default",
+				},
 			},
 		},
 		{
-			node: newNode("node0"),
+			// User specified node as schedulable and kubelet /healthz probe returns failure with no error.
+			// Expected node condition to be not ready and marked schedulable.
+			node: &api.Node{ObjectMeta: api.ObjectMeta{Name: "node0"}, Spec: api.NodeSpec{Unschedulable: false}},
 			fakeKubeletClient: &FakeKubeletClient{
 				Status: probe.Failure,
 				Err:    nil,
@@ -582,10 +591,17 @@ func TestHealthCheckNode(t *testing.T) {
 					Status: api.ConditionNone,
 					Reason: "Node health check failed: kubelet /healthz endpoint returns not ok",
 				},
+				{
+					Type:   api.NodeSchedulable,
+					Status: api.ConditionFull,
+					Reason: "Node is schedulable by default",
+				},
 			},
 		},
 		{
-			node: newNode("node0"),
+			// User specified node as unschedulable and kubelet /healthz probe returns failure with some error.
+			// Expected node condition to be not ready and marked unschedulable.
+			node: &api.Node{ObjectMeta: api.ObjectMeta{Name: "node0"}, Spec: api.NodeSpec{Unschedulable: true}},
 			fakeKubeletClient: &FakeKubeletClient{
 				Status: probe.Failure,
 				Err:    errors.New("Error"),
@@ -596,6 +612,11 @@ func TestHealthCheckNode(t *testing.T) {
 					Status: api.ConditionUnknown,
 					Reason: "Node health check error: Error",
 				},
+				{
+					Type:   api.NodeSchedulable,
+					Status: api.ConditionNone,
+					Reason: "User marked unschedulable during node create/update",
+				},
 			},
 		},
 	}
@@ -663,11 +684,13 @@ func TestSyncNodeStatusTransitionTime(t *testing.T) {
 		expectedTransitionTimeChange bool
 	}{
 		{
-			// Existing node is healthy, current porbe is healthy too.
+			// Existing node is healthy, current probe is healthy too.
+			// Existing node is schedulable, again explicitly mark node as schedulable.
 			fakeNodeHandler: &FakeNodeHandler{
 				Existing: []*api.Node{
 					{
 						ObjectMeta: api.ObjectMeta{Name: "node0"},
+						Spec:       api.NodeSpec{Unschedulable: false},
 						Status: api.NodeStatus{
 							Conditions: []api.NodeCondition{
 								{
@@ -676,6 +699,12 @@ func TestSyncNodeStatusTransitionTime(t *testing.T) {
 									Reason:             "Node health check succeeded: kubelet /healthz endpoint returns ok",
 									LastTransitionTime: util.Date(2012, 1, 1, 0, 0, 0, 0, time.UTC),
 								},
+								{
+									Type:               api.NodeSchedulable,
+									Status:             api.ConditionFull,
+									Reason:             "Node is schedulable by default",
+									LastTransitionTime: util.Date(2012, 1, 1, 0, 0, 0, 0, time.UTC),
+								},
 							},
 						},
 					},
@@ -689,11 +718,13 @@ func TestSyncNodeStatusTransitionTime(t *testing.T) {
 			expectedTransitionTimeChange: false,
 		},
 		{
-			// Existing node is healthy, current porbe is unhealthy.
+			// Existing node is healthy, current probe is unhealthy.
+			// Existing node is schedulable, mark node as unschedulable.
 			fakeNodeHandler: &FakeNodeHandler{
 				Existing: []*api.Node{
 					{
 						ObjectMeta: api.ObjectMeta{Name: "node0"},
+						Spec:       api.NodeSpec{Unschedulable: true},
 						Status: api.NodeStatus{
 							Conditions: []api.NodeCondition{
 								{
@@ -702,6 +733,12 @@ func TestSyncNodeStatusTransitionTime(t *testing.T) {
 									Reason:             "Node health check succeeded: kubelet /healthz endpoint returns ok",
 									LastTransitionTime: util.Date(2012, 1, 1, 0, 0, 0, 0, time.UTC),
 								},
+								{
+									Type:               api.NodeSchedulable,
+									Status:             api.ConditionFull,
+									Reason:             "Node is schedulable by default",
+									LastTransitionTime: util.Date(2012, 1, 1, 0, 0, 0, 0, time.UTC),
+								},
 							},
 						},
 					},
@@ -865,7 +902,7 @@ func TestSyncNodeStatusDeletePods(t *testing.T) {
 		expectedActions      []client.FakeAction
 	}{
 		{
-			// Existing node is healthy, current porbe is healthy too.
+			// Existing node is healthy, current probe is healthy too.
 			fakeNodeHandler: &FakeNodeHandler{
 				Existing: []*api.Node{
 					{
@@ -894,7 +931,7 @@ func TestSyncNodeStatusDeletePods(t *testing.T) {
 			expectedActions:      nil,
 		},
 		{
-			// Existing node is healthy, current porbe is unhealthy, i.e. node just becomes unhealthy.
+			// Existing node is healthy, current probe is unhealthy, i.e. node just becomes unhealthy.
 			// Do not delete pods.
 			fakeNodeHandler: &FakeNodeHandler{
 				Existing: []*api.Node{
@@ -924,7 +961,7 @@ func TestSyncNodeStatusDeletePods(t *testing.T) {
 			expectedActions:      nil,
 		},
 		{
-			// Existing node unhealthy, current porbe is unhealthy. Node is still within grace peroid.
+			// Existing node unhealthy, current probe is unhealthy. Node is still within grace peroid.
 			fakeNodeHandler: &FakeNodeHandler{
 				Existing: []*api.Node{
 					{
@@ -956,7 +993,7 @@ func TestSyncNodeStatusDeletePods(t *testing.T) {
 			expectedActions:      nil,
 		},
 		{
-			// Existing node unhealthy, current porbe is unhealthy. Node exceeds grace peroid.
+			// Existing node unhealthy, current probe is unhealthy. Node exceeds grace peroid.
 			fakeNodeHandler: &FakeNodeHandler{
 				Existing: []*api.Node{
 					{
@@ -1036,6 +1073,11 @@ func TestSyncNodeStatus(t *testing.T) {
 								Status: api.ConditionFull,
 								Reason: "Node health check succeeded: kubelet /healthz endpoint returns ok",
 							},
+							{
+								Type:   api.NodeSchedulable,
+								Status: api.ConditionFull,
+								Reason: "Node is schedulable by default",
+							},
 						},
 						Addresses: []api.NodeAddress{
 							{Type: api.NodeLegacyHostIP, Address: "1.2.3.4"},
@@ -1051,6 +1093,11 @@ func TestSyncNodeStatus(t *testing.T) {
 								Status: api.ConditionFull,
 								Reason: "Node health check succeeded: kubelet /healthz endpoint returns ok",
 							},
+							{
+								Type:   api.NodeSchedulable,
+								Status: api.ConditionFull,
+								Reason: "Node is schedulable by default",
+							},
 						},
 						Addresses: []api.NodeAddress{
 							{Type: api.NodeLegacyHostIP, Address: "1.2.3.4"},

pkg/kubectl/resource_printer.go

@@ -432,7 +432,7 @@ func printSecretList(list *api.SecretList, w io.Writer) error {
 
 func printNode(node *api.Node, w io.Writer) error {
 	conditionMap := make(map[api.NodeConditionType]*api.NodeCondition)
-	NodeAllConditions := []api.NodeConditionType{api.NodeReady, api.NodeReachable}
+	NodeAllConditions := []api.NodeConditionType{api.NodeSchedulable, api.NodeReady, api.NodeReachable}
 	for i := range node.Status.Conditions {
 		cond := node.Status.Conditions[i]
 		conditionMap[cond.Type] = &cond

pkg/kubectl/resource_printer_test.go

@@ -567,6 +567,26 @@ func TestPrintMinionStatus(t *testing.T) {
 			},
 			status: "Unknown",
 		},
+		{
+			minion: api.Node{
+				ObjectMeta: api.ObjectMeta{Name: "foo7"},
+				Status: api.NodeStatus{Conditions: []api.NodeCondition{
+					{Type: api.NodeSchedulable, Status: api.ConditionFull},
+					{Type: api.NodeReady, Status: api.ConditionFull},
+					{Type: api.NodeReachable, Status: api.ConditionFull}}},
+			},
+			status: "Schedulable,Ready,Reachable",
+		},
+		{
+			minion: api.Node{
+				ObjectMeta: api.ObjectMeta{Name: "foo8"},
+				Status: api.NodeStatus{Conditions: []api.NodeCondition{
+					{Type: api.NodeSchedulable, Status: api.ConditionNone},
+					{Type: api.NodeReady, Status: api.ConditionNone},
+					{Type: api.NodeReachable, Status: api.ConditionFull}}},
+			},
+			status: "NotSchedulable,NotReady,Reachable",
+		},
 	}
 
 	for _, test := range table {

pkg/registry/minion/rest.go

@@ -46,7 +46,6 @@ func NewREST(m Registry) *REST {
 }
 
 var ErrDoesNotExist = errors.New("The requested resource does not exist.")
-var ErrNotHealty = errors.New("The requested minion is not healthy.")
 
 // Create satisfies the RESTStorage interface.
 func (rs *REST) Create(ctx api.Context, obj runtime.Object) (runtime.Object, error) {

plugin/pkg/scheduler/factory/factory.go

@@ -207,6 +207,11 @@ func (factory *ConfigFactory) pollMinions() (cache.Enumerator, error) {
 			cond := node.Status.Conditions[i]
 			conditionMap[cond.Type] = &cond
 		}
+		if condition, ok := conditionMap[api.NodeSchedulable]; ok {
+			if condition.Status != api.ConditionFull {
+				continue
+			}
+		}
 		if condition, ok := conditionMap[api.NodeReady]; ok {
 			if condition.Status == api.ConditionFull {
 				nodes.Items = append(nodes.Items, node)

plugin/pkg/scheduler/factory/factory_test.go

@@ -156,7 +156,15 @@ func TestPollMinions(t *testing.T) {
 					},
 				},
 				{
-					ObjectMeta: api.ObjectMeta{Name: "baz"},
+					ObjectMeta: api.ObjectMeta{Name: "fiz"},
+					Status: api.NodeStatus{
+						Conditions: []api.NodeCondition{
+							{Type: api.NodeSchedulable, Status: api.ConditionFull},
+						},
+					},
+				},
+				{
+					ObjectMeta: api.ObjectMeta{Name: "biz"},
 					Status: api.NodeStatus{
 						Conditions: []api.NodeCondition{
 							{Type: api.NodeReady, Status: api.ConditionFull},
@@ -173,8 +181,47 @@ func TestPollMinions(t *testing.T) {
 						},
 					},
 				},
+				{
+					ObjectMeta: api.ObjectMeta{Name: "fuz"},
+					Status: api.NodeStatus{
+						Conditions: []api.NodeCondition{
+							{Type: api.NodeSchedulable, Status: api.ConditionFull},
+							{Type: api.NodeReady, Status: api.ConditionFull},
+							{Type: api.NodeReachable, Status: api.ConditionFull},
+						},
+					},
+				},
+				{
+					ObjectMeta: api.ObjectMeta{Name: "buz"},
+					Status: api.NodeStatus{
+						Conditions: []api.NodeCondition{
+							{Type: api.NodeSchedulable, Status: api.ConditionNone},
+							{Type: api.NodeReady, Status: api.ConditionFull},
+							{Type: api.NodeReachable, Status: api.ConditionFull},
+						},
+					},
+				},
+				{
+					ObjectMeta: api.ObjectMeta{Name: "foobar"},
+					Status: api.NodeStatus{
+						Conditions: []api.NodeCondition{
+							{Type: api.NodeSchedulable, Status: api.ConditionFull},
+							{Type: api.NodeReady, Status: api.ConditionNone},
+							{Type: api.NodeReachable, Status: api.ConditionFull},
+						},
+					},
+				},
+				{
+					ObjectMeta: api.ObjectMeta{Name: "fizbiz"},
+					Status: api.NodeStatus{
+						Conditions: []api.NodeCondition{
+							{Type: api.NodeSchedulable, Status: api.ConditionFull},
+							{Type: api.NodeReachable, Status: api.ConditionNone},
+						},
+					},
+				},
 			},
-			expectedCount: 4,
+			expectedCount: 6,
 		},
 		{
 			minions: []api.Node{
@@ -203,6 +250,27 @@ func TestPollMinions(t *testing.T) {
 					ObjectMeta: api.ObjectMeta{Name: "foo"},
 					Status: api.NodeStatus{
 						Conditions: []api.NodeCondition{
+							{Type: api.NodeSchedulable, Status: api.ConditionFull},
+						},
+					},
+				},
+				{
+					ObjectMeta: api.ObjectMeta{Name: "bar"},
+					Status: api.NodeStatus{
+						Conditions: []api.NodeCondition{
+							{Type: api.NodeSchedulable, Status: api.ConditionNone},
+						},
+					},
+				},
+			},
+			expectedCount: 1,
+		},
+		{
+			minions: []api.Node{
+				{
+					ObjectMeta: api.ObjectMeta{Name: "foo"},
+					Status: api.NodeStatus{
+						Conditions: []api.NodeCondition{
 							{Type: api.NodeReady, Status: api.ConditionFull},
 							{Type: api.NodeReachable, Status: api.ConditionNone}},
 					},