Max-allowed Pods for nodes

pkg/firmament/resource_desc.proto

@@ -80,4 +80,6 @@ message ResourceDescriptor {
   uint64 trace_machine_id = 21;
   // Resource labels
   repeated Label labels = 32;
+  // Max pods allowed per node
+  uint64 max_pods = 33;
 }

pkg/k8sclient/nodewatcher.go

@@ -106,6 +106,7 @@ func (nw *NodeWatcher) parseNode(node *v1.Node, phase NodePhase) *Node {
 	memCap, _ := memCapQuantity.AsInt64()
 	memAllocQuantity := node.Status.Allocatable[v1.ResourceMemory]
 	memAlloc, _ := memAllocQuantity.AsInt64()
+	podAllocQuantity := node.Status.Allocatable[v1.ResourcePods]
 	return &Node{
 		Hostname:         node.Name,
 		Phase:            phase,
@@ -115,6 +116,7 @@ func (nw *NodeWatcher) parseNode(node *v1.Node, phase NodePhase) *Node {
 		CPUAllocatable:   cpuAllocQuantity.MilliValue(),
 		MemCapacityKb:    memCap / bytesToKb,
 		MemAllocatableKb: memAlloc / bytesToKb,
+		PodAllocatable:   podAllocQuantity.Value(),
 		Labels:           node.Labels,
 		Annotations:      node.Annotations,
 	}
@@ -302,6 +304,7 @@ func (nw *NodeWatcher) createResourceTopologyForNode(node *Node) *firmament.Reso
 				RamCap:   uint64(node.MemCapacityKb),
 				CpuCores: float32(node.CPUCapacity),
 			},
+			MaxPods: uint64(node.PodAllocatable),
 		},
 	}
 	ResIDToNode[resUUID] = node.Hostname

pkg/k8sclient/types.go

@@ -71,6 +71,7 @@ type Node struct {
 	CPUAllocatable   int64
 	MemCapacityKb    int64
 	MemAllocatableKb int64
+	PodAllocatable   int64
 	Labels           map[string]string
 	Annotations      map[string]string
 }

test/e2e/framework/framework.go

@@ -164,7 +164,7 @@ func (f *Framework) AfterEach() {
 	f.FetchLogsFromFirmament(f.TestingNS)
 	f.FetchLogsFromPoseidon(f.TestingNS)
 	Logf("Delete namespace called")
-	err = f.deleteNamespace(f.TestingNS)
+	err = f.DeleteNamespace(f.TestingNS)
 	Expect(err).NotTo(HaveOccurred())
 
 	err = f.DeleteDeploymentIfExist(f.TestingNS, poseidonDeploymentName)

test/e2e/framework/namespace.go

@@ -62,14 +62,14 @@ func (f *Framework) deleteNamespaceIfExist(nsName string) error {
 	} else {
 		//delete the namespace as it exist
 		Logf("Deleting %v namespace as it exists", nsName)
-		if err = f.deleteNamespace(nsName); err != nil {
+		if err = f.DeleteNamespace(nsName); err != nil {
 			return fmt.Errorf("unable to delete %v namespace, error: %v occurred", nsName, err)
 		}
 	}
 	return nil
 }
 
-func (f *Framework) deleteNamespace(nsName string) error {
+func (f *Framework) DeleteNamespace(nsName string) error {
 
 	if err := f.ClientSet.CoreV1().Namespaces().Delete(nsName, nil); err != nil {
 		return err

test/e2e/poseidon_integration.go

@@ -20,6 +20,8 @@ import (
 	"fmt"
 	"math/rand"
 	"os"
+	"strconv"
+	"time"
 
 	"github.com/golang/glog"
 	"github.com/kubernetes-sigs/poseidon/test/e2e/framework"
@@ -32,7 +34,6 @@ import (
 	"k8s.io/apimachinery/pkg/api/resource"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/client-go/kubernetes"
-	"time"
 )
 
 var _ = Describe("Poseidon", func() {
@@ -483,6 +484,91 @@ var _ = Describe("Poseidon", func() {
 			Expect(err).To(HaveOccurred())
 		})
 
+		// Test whether the kubelet flag max_pods works
+		// Currently the test uses the node's default max_pods value, i.e., we don't manually set the kubelet flag max_pods;
+		// And only test on one of the nodes.
+		It("validates MaxPods limit number of pods that are allowed to run [Slow]", func() {
+
+			nodeList, err := clientset.CoreV1().Nodes().List(metav1.ListOptions{})
+			Expect(err).NotTo(HaveOccurred())
+			maxPodNamespace := "max-pods-test"
+			_, err = clientset.CoreV1().Namespaces().Create(&v1.Namespace{
+				ObjectMeta: metav1.ObjectMeta{Name: maxPodNamespace},
+			})
+			if err != nil {
+				glog.Errorf("unable to create namespace %s, error: %v occurred", maxPodNamespace, err)
+			}
+			for idx, node := range nodeList.Items {
+				framework.Logf("Node: %v", node.Name)
+				podAlloc, found := node.Status.Allocatable[v1.ResourcePods]
+				Expect(found).To(Equal(true))
+				podList, err := clientset.CoreV1().Pods(metav1.NamespaceAll).List(metav1.ListOptions{})
+				Expect(err).NotTo(HaveOccurred())
+				podCount := podList.Size()
+
+				name := "max-pods-" + strconv.Itoa(idx)
+				var replicas int32
+				replicas = int32(podAlloc.Value())
+				if podCount < 1 {
+					replicas += 5
+				}
+				_, err = clientset.ExtensionsV1beta1().Deployments(maxPodNamespace).Create(&v1beta1.Deployment{
+					ObjectMeta: metav1.ObjectMeta{
+						Labels: map[string]string{"app": "nginx"},
+						Name:   name,
+					},
+					Spec: v1beta1.DeploymentSpec{
+						Selector: &metav1.LabelSelector{
+							MatchLabels: map[string]string{"app": "nginx", "name": "test-max-pods"},
+						},
+						Replicas: &replicas,
+						Template: v1.PodTemplateSpec{
+							ObjectMeta: metav1.ObjectMeta{
+								Labels: map[string]string{"name": "test-max-pods", "app": "nginx", "schedulerName": "poseidon"},
+								Name:   name,
+							},
+							Spec: v1.PodSpec{
+								SchedulerName: "poseidon",
+								Containers: []v1.Container{
+									{
+										Name:            fmt.Sprintf("container-%s", name),
+										Image:           "nginx:latest",
+										ImagePullPolicy: "IfNotPresent",
+									},
+								},
+							},
+						},
+					},
+				})
+				Expect(err).NotTo(HaveOccurred())
+				// time for pods to be Running
+				time.Sleep(3 * time.Minute)
+				podList, err = clientset.CoreV1().Pods(metav1.NamespaceAll).List(metav1.ListOptions{})
+				Expect(err).NotTo(HaveOccurred())
+				var podRunningCount, podPendingCount, podFailedCount int64 = 0, 0, 0
+				for _, pod := range podList.Items {
+					if pod.Status.Phase == v1.PodRunning {
+						podRunningCount++
+					} else if pod.Status.Phase == v1.PodPending {
+						podPendingCount++
+					} else if pod.Status.Phase == v1.PodFailed {
+						podFailedCount++
+					}
+				}
+				Expect(podRunningCount == podAlloc.Value()).To(Equal(true))
+				glog.Errorf("Pods: %d Running, %d Pening, %d Failed", podRunningCount, podPendingCount, podFailedCount)
+
+				framework.Logf("Time to clean up the ns [%s] now...", maxPodNamespace)
+				if err = f.DeleteNamespace(maxPodNamespace); err != nil {
+					glog.Errorf("unable to delete %v namespace, error: %v occurred", maxPodNamespace, err)
+				}
+				// time for Terminating pods to disappear
+				// TODO: deal with the deletion better. Tried deleting the deployment but didn't work well.
+				time.Sleep(1 * time.Minute)
+				// Test one node for now since the test takes much time.
+				break
+			}
+		})
 		// Test Nodes does not have any label, hence it should be impossible to schedule Pod with
 		// nonempty Selector set.
 		/*