Merge pull request #20 from ravisantoshgudimetla/pod-anti-affinity-strategy

Pod antiaffinity strategy

examples/policy.yaml

@@ -3,6 +3,8 @@ kind: "DeschedulerPolicy"
 strategies:
   "RemoveDuplicates":
      enabled: true
+  "RemovePodsViolatingInterPodAntiAffinity":
+     enabled: true
   "LowNodeUtilization":
      enabled: true
      params:

pkg/descheduler/descheduler.go

@@ -55,6 +55,7 @@ func Run(rs *options.DeschedulerServer) error {
 
 	strategies.RemoveDuplicatePods(rs, deschedulerPolicy.Strategies["RemoveDuplicates"], evictionPolicyGroupVersion, nodes)
 	strategies.LowNodeUtilization(rs, deschedulerPolicy.Strategies["LowNodeUtilization"], evictionPolicyGroupVersion, nodes)
+	strategies.RemovePodsViolatingInterPodAntiAffinity(rs, deschedulerPolicy.Strategies["RemovePodsViolatingInterPodAntiAffinity"], evictionPolicyGroupVersion, nodes)
 
 	return nil
 }

pkg/descheduler/strategies/pod_antiaffinity.go

@@ -0,0 +1,102 @@
+/*
+Copyright 2017 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 strategies
+
+import (
+	"github.com/kubernetes-incubator/descheduler/cmd/descheduler/app/options"
+	"github.com/kubernetes-incubator/descheduler/pkg/api"
+	"k8s.io/kubernetes/pkg/api/v1"
+
+	"github.com/golang/glog"
+	"github.com/kubernetes-incubator/descheduler/pkg/descheduler/evictions"
+	podutil "github.com/kubernetes-incubator/descheduler/pkg/descheduler/pod"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	//TODO: Change to client-go instead of generated clientset.
+	"k8s.io/kubernetes/pkg/client/clientset_generated/clientset"
+	priorityutil "k8s.io/kubernetes/plugin/pkg/scheduler/algorithm/priorities/util"
+	"k8s.io/kubernetes/plugin/pkg/scheduler/schedulercache"
+)
+
+// RemovePodsViolatingInterPodAntiAffinity with elimination strategy
+func RemovePodsViolatingInterPodAntiAffinity(ds *options.DeschedulerServer, strategy api.DeschedulerStrategy, policyGroupVersion string, nodes []*v1.Node) {
+	if !strategy.Enabled {
+		return
+	}
+	removePodsWithAffinityRules(ds.Client, policyGroupVersion, nodes, ds.DryRun)
+}
+
+// removePodsWithAffinityRules evicts pods on the node which are having a pod affinity rules.
+func removePodsWithAffinityRules(client clientset.Interface, policyGroupVersion string, nodes []*v1.Node, dryRun bool) int {
+	podsEvicted := 0
+	for _, node := range nodes {
+		glog.V(1).Infof("Processing node: %#v\n", node.Name)
+		pods, err := podutil.ListPodsOnANode(client, node)
+		if err != nil {
+			return 0
+		}
+		totalPods := len(pods)
+		for i := 0; i < totalPods; i++ {
+			if checkPodsWithAntiAffinityExist(pods[i], pods) {
+				success, err := evictions.EvictPod(client, pods[i], policyGroupVersion, dryRun)
+				if !success {
+					glog.Infof("Error when evicting pod: %#v (%#v)\n", pods[i].Name, err)
+				} else {
+					podsEvicted++
+					glog.V(1).Infof("Evicted pod: %#v (%#v)\n because of existing anti-affinity", pods[i].Name, err)
+					// Since the current pod is evicted all other pods which have anti-affinity with this
+					// pod need not be evicted.
+					// Update pods.
+					pods = append(pods[:i], pods[i+1:]...)
+					i--
+					totalPods--
+				}
+			}
+		}
+	}
+	return podsEvicted
+}
+
+// checkPodsWithAntiAffinityExist checks if there are other pods on the node that the current pod cannot tolerate.
+func checkPodsWithAntiAffinityExist(pod *v1.Pod, pods []*v1.Pod) bool {
+	affinity := schedulercache.ReconcileAffinity(pod)
+	if affinity != nil && affinity.PodAntiAffinity != nil {
+		for _, term := range getPodAntiAffinityTerms(affinity.PodAntiAffinity) {
+			namespaces := priorityutil.GetNamespacesFromPodAffinityTerm(pod, &term)
+			selector, err := metav1.LabelSelectorAsSelector(term.LabelSelector)
+			if err != nil {
+				glog.Infof("%v", err)
+				return false
+			}
+			for _, existingPod := range pods {
+				if existingPod.Name != pod.Name && priorityutil.PodMatchesTermsNamespaceAndSelector(existingPod, namespaces, selector) {
+					return true
+				}
+			}
+		}
+	}
+	return false
+}
+
+// getPodAntiAffinityTerms gets the antiaffinity terms for the given pod.
+func getPodAntiAffinityTerms(podAntiAffinity *v1.PodAntiAffinity) (terms []v1.PodAffinityTerm) {
+	if podAntiAffinity != nil {
+		if len(podAntiAffinity.RequiredDuringSchedulingIgnoredDuringExecution) != 0 {
+			terms = podAntiAffinity.RequiredDuringSchedulingIgnoredDuringExecution
+		}
+	}
+	return terms
+}

pkg/descheduler/strategies/pod_antiaffinity_test.go

@@ -0,0 +1,86 @@
+/*
+Copyright 2017 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 strategies
+
+import (
+	"testing"
+
+	"fmt"
+	"github.com/kubernetes-incubator/descheduler/test"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/runtime"
+	core "k8s.io/client-go/testing"
+	"k8s.io/kubernetes/pkg/api/v1"
+	"k8s.io/kubernetes/pkg/client/clientset_generated/clientset/fake"
+)
+
+func TestPodAntiAffinity(t *testing.T) {
+	node := test.BuildTestNode("n1", 2000, 3000, 10)
+	p1 := test.BuildTestPod("p1", 100, 0, node.Name)
+	p2 := test.BuildTestPod("p2", 100, 0, node.Name)
+	p3 := test.BuildTestPod("p3", 100, 0, node.Name)
+	p3.Labels = map[string]string{"foo": "bar"}
+	p1.Spec.Affinity = &v1.Affinity{
+		PodAntiAffinity: &v1.PodAntiAffinity{
+			RequiredDuringSchedulingIgnoredDuringExecution: []v1.PodAffinityTerm{
+				{
+					LabelSelector: &metav1.LabelSelector{
+						MatchExpressions: []metav1.LabelSelectorRequirement{
+							{
+								Key:      "foo",
+								Operator: metav1.LabelSelectorOpIn,
+								Values:   []string{"bar"},
+							},
+						},
+					},
+					TopologyKey: "region",
+				},
+			},
+		},
+	}
+	p3.Spec.Affinity = &v1.Affinity{
+		PodAntiAffinity: &v1.PodAntiAffinity{
+			RequiredDuringSchedulingIgnoredDuringExecution: []v1.PodAffinityTerm{
+				{
+					LabelSelector: &metav1.LabelSelector{
+						MatchExpressions: []metav1.LabelSelectorRequirement{
+							{
+								Key:      "foo",
+								Operator: metav1.LabelSelectorOpIn,
+								Values:   []string{"bar"},
+							},
+						},
+					},
+					TopologyKey: "region",
+				},
+			},
+		},
+	}
+	fakeClient := &fake.Clientset{}
+	fakeClient.Fake.AddReactor("list", "pods", func(action core.Action) (bool, runtime.Object, error) {
+		return true, &v1.PodList{Items: []v1.Pod{*p1, *p2, *p3}}, nil
+	})
+	fakeClient.Fake.AddReactor("get", "nodes", func(action core.Action) (bool, runtime.Object, error) {
+		return true, node, nil
+	})
+	expectedEvictedPodCount := 1
+	podsEvicted := removePodsWithAffinityRules(fakeClient, "v1", []*v1.Node{node}, false)
+	if podsEvicted != expectedEvictedPodCount {
+		fmt.Println(podsEvicted)
+		t.Errorf("Unexpected no of pods evicted")
+	}
+}