Merge pull request #142 from adyada/master

Added support to evict pods that use local storage.

CONTRIBUTING.md

@@ -0,0 +1,23 @@
+# Contributing Guidelines
+
+Welcome to Kubernetes. We are excited about the prospect of you joining our [community](https://github.com/kubernetes/community)! The Kubernetes community abides by the CNCF [code of conduct](code-of-conduct.md). Here is an excerpt:
+
+_As contributors and maintainers of this project, and in the interest of fostering an open and welcoming community, we pledge to respect all people who contribute through reporting issues, posting feature requests, updating documentation, submitting pull requests or patches, and other activities._
+
+## Getting Started
+
+We have full documentation on how to get started contributing here:
+
+- [Contributor License Agreement](https://git.k8s.io/community/CLA.md) Kubernetes projects require that you sign a Contributor License Agreement (CLA) before we can accept your pull requests
+- [Kubernetes Contributor Guide](http://git.k8s.io/community/contributors/guide) - Main contributor documentation, or you can just jump directly to the [contributing section](http://git.k8s.io/community/contributors/guide#contributing)
+- [Contributor Cheat Sheet](https://git.k8s.io/community/contributors/guide/contributor-cheatsheet.md) - Common resources for existing developers
+
+## Mentorship
+
+- [Mentoring Initiatives](https://git.k8s.io/community/mentoring) - We have a diverse set of mentorship programs available that are always looking for volunteers!
+
+
+## Contact Information
+
+- [Slack channel](https://kubernetes.slack.com/messages/sig-scheduling)
+- [Mailing list](https://groups.google.com/forum/#!forum/kubernetes-sig-scheduling)

README.md

@@ -138,11 +138,7 @@ spec:
           volumeMounts:
           - mountPath: /policy-dir
             name: policy-volume
-          command:
+          command: ["/bin/descheduler",  "--policy-config-file", "/policy-dir/policy.yaml"]
-          - "/bin/sh"
-          - "-ec"
-          - |
-            /bin/descheduler --policy-config-file /policy-dir/policy.yaml
         restartPolicy: "Never"
         serviceAccountName: descheduler-sa
         volumes:
@@ -285,11 +281,18 @@ This roadmap is not in any particular order.
 
 Descheduler  | supported Kubernetes version
 -------------|-----------------------------
-0.4          | 1.9+
+0.4+          | 1.9+
 0.1-0.3      | 1.7-1.8
 
-## Note
+## Community, discussion, contribution, and support
 
-This project is under active development, and is not intended for production use.
+Learn how to engage with the Kubernetes community on the [community page](http://kubernetes.io/community/).
-Any api could be changed any time with out any notice. That said, your feedback is
+
-very important and appreciated to make this project more stable and useful.
+You can reach the maintainers of this project at:
+
+- [Slack channel](https://kubernetes.slack.com/messages/sig-scheduling)
+- [Mailing list](https://groups.google.com/forum/#!forum/kubernetes-sig-scheduling)
+
+### Code of conduct
+
+Participation in the Kubernetes community is governed by the [Kubernetes Code of Conduct](code-of-conduct.md).

cmd/descheduler/app/options/options.go

@@ -57,4 +57,6 @@ func (rs *DeschedulerServer) AddFlags(fs *pflag.FlagSet) {
 	fs.StringVar(&rs.NodeSelector, "node-selector", rs.NodeSelector, "Selector (label query) to filter on, supports '=', '==', and '!='.(e.g. -l key1=value1,key2=value2)")
 	// max-no-pods-to-evict limits the maximum number of pods to be evicted per node by descheduler.
 	fs.IntVar(&rs.MaxNoOfPodsToEvictPerNode, "max-pods-to-evict-per-node", rs.MaxNoOfPodsToEvictPerNode, "Limits the maximum number of pods to be evicted per node by descheduler")
+	// evict-local-storage-pods allows eviction of pods that are using local storage. This is false by default.
+	fs.BoolVar(&rs.EvictLocalStoragePods, "evict-local-storage-pods", rs.EvictLocalStoragePods, "Enables evicting pods using local storage by descheduler")
 }

hack/e2e-gce/gcloud_create_cluster.sh

@@ -10,7 +10,7 @@ master_uuid=$(uuid)
 node1_uuid=$(uuid)
 node2_uuid=$(uuid)
 kube_apiserver_port=6443
-kube_version=1.9.4
+kube_version=1.13.1
 
 DESCHEDULER_ROOT=$(dirname "${BASH_SOURCE}")/../../
 E2E_GCE_HOME=$DESCHEDULER_ROOT/hack/e2e-gce
@@ -36,10 +36,10 @@ create_cluster() {
 
 generate_kubeadm_instance_files() {
 	# TODO: Check if they have come up. awk $6 contains the state(RUNNING or not).
-	master_public_ip=$(gcloud compute instances list | grep $master_uuid|awk '{print $5}')
+	master_private_ip=$(gcloud compute instances list | grep $master_uuid|awk '{print $4}')
 	node1_public_ip=$(gcloud compute instances list | grep $node1_uuid|awk '{print $5}')
 	node2_public_ip=$(gcloud compute instances list | grep $node2_uuid|awk '{print $5}')
-	echo "kubeadm init --kubernetes-version=${kube_version} --apiserver-advertise-address=${master_public_ip}" --skip-preflight-checks --pod-network-cidr=10.96.0.0/12 > $E2E_GCE_HOME/kubeadm_install.sh
+	echo "kubeadm init --kubernetes-version=${kube_version} --apiserver-advertise-address=${master_private_ip}" --ignore-preflight-errors=all --pod-network-cidr=10.96.0.0/12 > $E2E_GCE_HOME/kubeadm_install.sh
 }
 
 

kubernetes/configmap.yaml

@@ -0,0 +1,27 @@
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: descheduler-policy-configmap
+  namespace: kube-system
+data:
+  policy.yaml: |
+    apiVersion: "descheduler/v1alpha1"
+    kind: "DeschedulerPolicy"
+    strategies:
+      "RemoveDuplicates":
+         enabled: true
+      "RemovePodsViolatingInterPodAntiAffinity":
+         enabled: true
+      "LowNodeUtilization":
+         enabled: true
+         params:
+           nodeResourceUtilizationThresholds:
+             thresholds:
+               "cpu" : 20
+               "memory": 20
+               "pods": 20
+             targetThresholds:
+               "cpu" : 50
+               "memory": 50
+               "pods": 50
+

kubernetes/job.yaml

@@ -0,0 +1,33 @@
+apiVersion: batch/v1
+kind: Job
+metadata:
+  name: descheduler-job
+  namespace: kube-system
+spec:
+  parallelism: 1
+  completions: 1
+  template:
+    metadata:
+      name: descheduler-pod
+      annotations:
+        scheduler.alpha.kubernetes.io/critical-pod: ""
+    spec:
+        containers:
+        - name: descheduler
+          image: descheduler:latest
+          volumeMounts:
+          - mountPath: /policy-dir
+            name: policy-volume
+          command:
+            - "/bin/descheduler"
+          args:
+            - "--policy-config-file"
+            - "/policy-dir/policy.yaml"
+            - "--v"
+            - "3"
+        restartPolicy: "Never"
+        serviceAccountName: descheduler-sa
+        volumes:
+        - name: policy-volume
+          configMap:
+            name: descheduler-policy-configmap

kubernetes/rbac.yaml

@@ -0,0 +1,37 @@
+---
+kind: ClusterRole
+apiVersion: rbac.authorization.k8s.io/v1
+metadata:
+  name: descheduler-cluster-role
+  namespace: kube-system
+rules:
+- apiGroups: [""]
+  resources: ["nodes"]
+  verbs: ["get", "watch", "list"]
+- apiGroups: [""]
+  resources: ["pods"]
+  verbs: ["get", "watch", "list", "delete"]
+- apiGroups: [""]
+  resources: ["pods/eviction"]
+  verbs: ["create"]
+---
+apiVersion: v1
+kind: ServiceAccount
+metadata:
+  name: descheduler-sa
+  namespace: kube-system
+---
+apiVersion: rbac.authorization.k8s.io/v1
+kind: ClusterRoleBinding
+metadata:
+  name: descehduler-cluster-role-binding
+  namespace: kube-system
+roleRef:
+  apiGroup: rbac.authorization.k8s.io
+  kind: ClusterRole
+  name: descheduler-cluster-role
+subjects:
+  - name: descheduler-sa
+    kind: ServiceAccount
+    namespace: kube-system
+

pkg/apis/componentconfig/types.generated.go

@@ -77,16 +77,16 @@ func (x *DeschedulerConfiguration) CodecEncodeSelf(e *codec1978.Encoder) {
 		} else {
 			yysep2 := !z.EncBinary()
 			yy2arr2 := z.EncBasicHandle().StructToArray
-			var yyq2 [8]bool
+			var yyq2 [9]bool
 			_ = yyq2
 			_, _ = yysep2, yy2arr2
 			const yyr2 bool = false
 			yyq2[0] = x.Kind != ""
 			yyq2[1] = x.APIVersion != ""
 			if yyr2 || yy2arr2 {
-				r.WriteArrayStart(8)
+				r.WriteArrayStart(9)
 			} else {
-				var yynn2 = 6
+				var yynn2 = 7
 				for _, b := range yyq2 {
 					if b {
 						yynn2++
@@ -263,6 +263,25 @@ func (x *DeschedulerConfiguration) CodecEncodeSelf(e *codec1978.Encoder) {
 					r.EncodeInt(int64(x.MaxNoOfPodsToEvictPerNode))
 				}
 			}
+			if yyr2 || yy2arr2 {
+				r.WriteArrayElem()
+				yym28 := z.EncBinary()
+				_ = yym28
+				if false {
+				} else {
+					r.EncodeBool(bool(x.EvictLocalStoragePods))
+				}
+			} else {
+				r.WriteMapElemKey()
+				r.EncStructFieldKey(codecSelferValueTypeString1234, `EvictLocalStoragePods`)
+				r.WriteMapElemValue()
+				yym29 := z.EncBinary()
+				_ = yym29
+				if false {
+				} else {
+					r.EncodeBool(bool(x.EvictLocalStoragePods))
+				}
+			}
 			if yyr2 || yy2arr2 {
 				r.WriteArrayEnd()
 			} else {
@@ -420,6 +439,18 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromMap(l int, d *codec1978.De
 					*((*int)(yyv18)) = int(r.DecodeInt(codecSelferBitsize1234))
 				}
 			}
+		case "EvictLocalStoragePods":
+			if r.TryDecodeAsNil() {
+				x.EvictLocalStoragePods = false
+			} else {
+				yyv20 := &x.EvictLocalStoragePods
+				yym21 := z.DecBinary()
+				_ = yym21
+				if false {
+				} else {
+					*((*bool)(yyv20)) = r.DecodeBool()
+				}
+			}
 		default:
 			z.DecStructFieldNotFound(-1, yys3)
 		} // end switch yys3
@@ -431,16 +462,16 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 	var h codecSelfer1234
 	z, r := codec1978.GenHelperDecoder(d)
 	_, _, _ = h, z, r
-	var yyj20 int
+	var yyj22 int
-	var yyb20 bool
+	var yyb22 bool
-	var yyhl20 bool = l >= 0
+	var yyhl22 bool = l >= 0
-	yyj20++
+	yyj22++
-	if yyhl20 {
+	if yyhl22 {
-		yyb20 = yyj20 > l
+		yyb22 = yyj22 > l
 	} else {
-		yyb20 = r.CheckBreak()
+		yyb22 = r.CheckBreak()
 	}
-	if yyb20 {
+	if yyb22 {
 		r.ReadArrayEnd()
 		return
 	}
@@ -448,29 +479,7 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 	if r.TryDecodeAsNil() {
 		x.Kind = ""
 	} else {
-		yyv21 := &x.Kind
+		yyv23 := &x.Kind
-		yym22 := z.DecBinary()
-		_ = yym22
-		if false {
-		} else {
-			*((*string)(yyv21)) = r.DecodeString()
-		}
-	}
-	yyj20++
-	if yyhl20 {
-		yyb20 = yyj20 > l
-	} else {
-		yyb20 = r.CheckBreak()
-	}
-	if yyb20 {
-		r.ReadArrayEnd()
-		return
-	}
-	r.ReadArrayElem()
-	if r.TryDecodeAsNil() {
-		x.APIVersion = ""
-	} else {
-		yyv23 := &x.APIVersion
 		yym24 := z.DecBinary()
 		_ = yym24
 		if false {
@@ -478,13 +487,35 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 			*((*string)(yyv23)) = r.DecodeString()
 		}
 	}
-	yyj20++
+	yyj22++
-	if yyhl20 {
+	if yyhl22 {
-		yyb20 = yyj20 > l
+		yyb22 = yyj22 > l
 	} else {
-		yyb20 = r.CheckBreak()
+		yyb22 = r.CheckBreak()
 	}
-	if yyb20 {
+	if yyb22 {
+		r.ReadArrayEnd()
+		return
+	}
+	r.ReadArrayElem()
+	if r.TryDecodeAsNil() {
+		x.APIVersion = ""
+	} else {
+		yyv25 := &x.APIVersion
+		yym26 := z.DecBinary()
+		_ = yym26
+		if false {
+		} else {
+			*((*string)(yyv25)) = r.DecodeString()
+		}
+	}
+	yyj22++
+	if yyhl22 {
+		yyb22 = yyj22 > l
+	} else {
+		yyb22 = r.CheckBreak()
+	}
+	if yyb22 {
 		r.ReadArrayEnd()
 		return
 	}
@@ -492,23 +523,23 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 	if r.TryDecodeAsNil() {
 		x.DeschedulingInterval = 0
 	} else {
-		yyv25 := &x.DeschedulingInterval
+		yyv27 := &x.DeschedulingInterval
-		yym26 := z.DecBinary()
+		yym28 := z.DecBinary()
-		_ = yym26
+		_ = yym28
 		if false {
-		} else if yyxt26 := z.Extension(z.I2Rtid(yyv25)); yyxt26 != nil {
+		} else if yyxt28 := z.Extension(z.I2Rtid(yyv27)); yyxt28 != nil {
-			z.DecExtension(yyv25, yyxt26)
+			z.DecExtension(yyv27, yyxt28)
 		} else {
-			*((*int64)(yyv25)) = int64(r.DecodeInt(64))
+			*((*int64)(yyv27)) = int64(r.DecodeInt(64))
 		}
 	}
-	yyj20++
+	yyj22++
-	if yyhl20 {
+	if yyhl22 {
-		yyb20 = yyj20 > l
+		yyb22 = yyj22 > l
 	} else {
-		yyb20 = r.CheckBreak()
+		yyb22 = r.CheckBreak()
 	}
-	if yyb20 {
+	if yyb22 {
 		r.ReadArrayEnd()
 		return
 	}
@@ -516,29 +547,7 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 	if r.TryDecodeAsNil() {
 		x.KubeconfigFile = ""
 	} else {
-		yyv27 := &x.KubeconfigFile
+		yyv29 := &x.KubeconfigFile
-		yym28 := z.DecBinary()
-		_ = yym28
-		if false {
-		} else {
-			*((*string)(yyv27)) = r.DecodeString()
-		}
-	}
-	yyj20++
-	if yyhl20 {
-		yyb20 = yyj20 > l
-	} else {
-		yyb20 = r.CheckBreak()
-	}
-	if yyb20 {
-		r.ReadArrayEnd()
-		return
-	}
-	r.ReadArrayElem()
-	if r.TryDecodeAsNil() {
-		x.PolicyConfigFile = ""
-	} else {
-		yyv29 := &x.PolicyConfigFile
 		yym30 := z.DecBinary()
 		_ = yym30
 		if false {
@@ -546,13 +555,35 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 			*((*string)(yyv29)) = r.DecodeString()
 		}
 	}
-	yyj20++
+	yyj22++
-	if yyhl20 {
+	if yyhl22 {
-		yyb20 = yyj20 > l
+		yyb22 = yyj22 > l
 	} else {
-		yyb20 = r.CheckBreak()
+		yyb22 = r.CheckBreak()
 	}
-	if yyb20 {
+	if yyb22 {
+		r.ReadArrayEnd()
+		return
+	}
+	r.ReadArrayElem()
+	if r.TryDecodeAsNil() {
+		x.PolicyConfigFile = ""
+	} else {
+		yyv31 := &x.PolicyConfigFile
+		yym32 := z.DecBinary()
+		_ = yym32
+		if false {
+		} else {
+			*((*string)(yyv31)) = r.DecodeString()
+		}
+	}
+	yyj22++
+	if yyhl22 {
+		yyb22 = yyj22 > l
+	} else {
+		yyb22 = r.CheckBreak()
+	}
+	if yyb22 {
 		r.ReadArrayEnd()
 		return
 	}
@@ -560,21 +591,21 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 	if r.TryDecodeAsNil() {
 		x.DryRun = false
 	} else {
-		yyv31 := &x.DryRun
+		yyv33 := &x.DryRun
-		yym32 := z.DecBinary()
+		yym34 := z.DecBinary()
-		_ = yym32
+		_ = yym34
 		if false {
 		} else {
-			*((*bool)(yyv31)) = r.DecodeBool()
+			*((*bool)(yyv33)) = r.DecodeBool()
 		}
 	}
-	yyj20++
+	yyj22++
-	if yyhl20 {
+	if yyhl22 {
-		yyb20 = yyj20 > l
+		yyb22 = yyj22 > l
 	} else {
-		yyb20 = r.CheckBreak()
+		yyb22 = r.CheckBreak()
 	}
-	if yyb20 {
+	if yyb22 {
 		r.ReadArrayEnd()
 		return
 	}
@@ -582,21 +613,21 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 	if r.TryDecodeAsNil() {
 		x.NodeSelector = ""
 	} else {
-		yyv33 := &x.NodeSelector
+		yyv35 := &x.NodeSelector
-		yym34 := z.DecBinary()
+		yym36 := z.DecBinary()
-		_ = yym34
+		_ = yym36
 		if false {
 		} else {
-			*((*string)(yyv33)) = r.DecodeString()
+			*((*string)(yyv35)) = r.DecodeString()
 		}
 	}
-	yyj20++
+	yyj22++
-	if yyhl20 {
+	if yyhl22 {
-		yyb20 = yyj20 > l
+		yyb22 = yyj22 > l
 	} else {
-		yyb20 = r.CheckBreak()
+		yyb22 = r.CheckBreak()
 	}
-	if yyb20 {
+	if yyb22 {
 		r.ReadArrayEnd()
 		return
 	}
@@ -604,26 +635,48 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 	if r.TryDecodeAsNil() {
 		x.MaxNoOfPodsToEvictPerNode = 0
 	} else {
-		yyv35 := &x.MaxNoOfPodsToEvictPerNode
+		yyv37 := &x.MaxNoOfPodsToEvictPerNode
-		yym36 := z.DecBinary()
+		yym38 := z.DecBinary()
-		_ = yym36
+		_ = yym38
 		if false {
 		} else {
-			*((*int)(yyv35)) = int(r.DecodeInt(codecSelferBitsize1234))
+			*((*int)(yyv37)) = int(r.DecodeInt(codecSelferBitsize1234))
+		}
+	}
+	yyj22++
+	if yyhl22 {
+		yyb22 = yyj22 > l
+	} else {
+		yyb22 = r.CheckBreak()
+	}
+	if yyb22 {
+		r.ReadArrayEnd()
+		return
+	}
+	r.ReadArrayElem()
+	if r.TryDecodeAsNil() {
+		x.EvictLocalStoragePods = false
+	} else {
+		yyv39 := &x.EvictLocalStoragePods
+		yym40 := z.DecBinary()
+		_ = yym40
+		if false {
+		} else {
+			*((*bool)(yyv39)) = r.DecodeBool()
 		}
 	}
 	for {
-		yyj20++
+		yyj22++
-		if yyhl20 {
+		if yyhl22 {
-			yyb20 = yyj20 > l
+			yyb22 = yyj22 > l
 		} else {
-			yyb20 = r.CheckBreak()
+			yyb22 = r.CheckBreak()
 		}
-		if yyb20 {
+		if yyb22 {
 			break
 		}
 		r.ReadArrayElem()
-		z.DecStructFieldNotFound(yyj20-1, "")
+		z.DecStructFieldNotFound(yyj22-1, "")
 	}
 	r.ReadArrayEnd()
 }

pkg/apis/componentconfig/types.go

@@ -45,4 +45,7 @@ type DeschedulerConfiguration struct {
 
 	// MaxNoOfPodsToEvictPerNode restricts maximum of pods to be evicted per node.
 	MaxNoOfPodsToEvictPerNode int
+
+	// EvictLocalStoragePods allows pods using local storage to be evicted.
+	EvictLocalStoragePods bool
 }

pkg/apis/componentconfig/v1alpha1/types.generated.go

@@ -77,7 +77,7 @@ func (x *DeschedulerConfiguration) CodecEncodeSelf(e *codec1978.Encoder) {
 		} else {
 			yysep2 := !z.EncBinary()
 			yy2arr2 := z.EncBasicHandle().StructToArray
-			var yyq2 [8]bool
+			var yyq2 [9]bool
 			_ = yyq2
 			_, _ = yysep2, yy2arr2
 			const yyr2 bool = false
@@ -88,8 +88,9 @@ func (x *DeschedulerConfiguration) CodecEncodeSelf(e *codec1978.Encoder) {
 			yyq2[5] = x.DryRun != false
 			yyq2[6] = x.NodeSelector != ""
 			yyq2[7] = x.MaxNoOfPodsToEvictPerNode != 0
+			yyq2[8] = x.EvictLocalStoragePods != false
 			if yyr2 || yy2arr2 {
-				r.WriteArrayStart(8)
+				r.WriteArrayStart(9)
 			} else {
 				var yynn2 = 1
 				for _, b := range yyq2 {
@@ -298,6 +299,31 @@ func (x *DeschedulerConfiguration) CodecEncodeSelf(e *codec1978.Encoder) {
 					}
 				}
 			}
+			if yyr2 || yy2arr2 {
+				r.WriteArrayElem()
+				if yyq2[8] {
+					yym28 := z.EncBinary()
+					_ = yym28
+					if false {
+					} else {
+						r.EncodeBool(bool(x.EvictLocalStoragePods))
+					}
+				} else {
+					r.EncodeBool(false)
+				}
+			} else {
+				if yyq2[8] {
+					r.WriteMapElemKey()
+					r.EncStructFieldKey(codecSelferValueTypeString1234, `evictLocalStoragePods`)
+					r.WriteMapElemValue()
+					yym29 := z.EncBinary()
+					_ = yym29
+					if false {
+					} else {
+						r.EncodeBool(bool(x.EvictLocalStoragePods))
+					}
+				}
+			}
 			if yyr2 || yy2arr2 {
 				r.WriteArrayEnd()
 			} else {
@@ -455,6 +481,18 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromMap(l int, d *codec1978.De
 					*((*int)(yyv18)) = int(r.DecodeInt(codecSelferBitsize1234))
 				}
 			}
+		case "evictLocalStoragePods":
+			if r.TryDecodeAsNil() {
+				x.EvictLocalStoragePods = false
+			} else {
+				yyv20 := &x.EvictLocalStoragePods
+				yym21 := z.DecBinary()
+				_ = yym21
+				if false {
+				} else {
+					*((*bool)(yyv20)) = r.DecodeBool()
+				}
+			}
 		default:
 			z.DecStructFieldNotFound(-1, yys3)
 		} // end switch yys3
@@ -466,16 +504,16 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 	var h codecSelfer1234
 	z, r := codec1978.GenHelperDecoder(d)
 	_, _, _ = h, z, r
-	var yyj20 int
+	var yyj22 int
-	var yyb20 bool
+	var yyb22 bool
-	var yyhl20 bool = l >= 0
+	var yyhl22 bool = l >= 0
-	yyj20++
+	yyj22++
-	if yyhl20 {
+	if yyhl22 {
-		yyb20 = yyj20 > l
+		yyb22 = yyj22 > l
 	} else {
-		yyb20 = r.CheckBreak()
+		yyb22 = r.CheckBreak()
 	}
-	if yyb20 {
+	if yyb22 {
 		r.ReadArrayEnd()
 		return
 	}
@@ -483,29 +521,7 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 	if r.TryDecodeAsNil() {
 		x.Kind = ""
 	} else {
-		yyv21 := &x.Kind
+		yyv23 := &x.Kind
-		yym22 := z.DecBinary()
-		_ = yym22
-		if false {
-		} else {
-			*((*string)(yyv21)) = r.DecodeString()
-		}
-	}
-	yyj20++
-	if yyhl20 {
-		yyb20 = yyj20 > l
-	} else {
-		yyb20 = r.CheckBreak()
-	}
-	if yyb20 {
-		r.ReadArrayEnd()
-		return
-	}
-	r.ReadArrayElem()
-	if r.TryDecodeAsNil() {
-		x.APIVersion = ""
-	} else {
-		yyv23 := &x.APIVersion
 		yym24 := z.DecBinary()
 		_ = yym24
 		if false {
@@ -513,13 +529,35 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 			*((*string)(yyv23)) = r.DecodeString()
 		}
 	}
-	yyj20++
+	yyj22++
-	if yyhl20 {
+	if yyhl22 {
-		yyb20 = yyj20 > l
+		yyb22 = yyj22 > l
 	} else {
-		yyb20 = r.CheckBreak()
+		yyb22 = r.CheckBreak()
 	}
-	if yyb20 {
+	if yyb22 {
+		r.ReadArrayEnd()
+		return
+	}
+	r.ReadArrayElem()
+	if r.TryDecodeAsNil() {
+		x.APIVersion = ""
+	} else {
+		yyv25 := &x.APIVersion
+		yym26 := z.DecBinary()
+		_ = yym26
+		if false {
+		} else {
+			*((*string)(yyv25)) = r.DecodeString()
+		}
+	}
+	yyj22++
+	if yyhl22 {
+		yyb22 = yyj22 > l
+	} else {
+		yyb22 = r.CheckBreak()
+	}
+	if yyb22 {
 		r.ReadArrayEnd()
 		return
 	}
@@ -527,23 +565,23 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 	if r.TryDecodeAsNil() {
 		x.DeschedulingInterval = 0
 	} else {
-		yyv25 := &x.DeschedulingInterval
+		yyv27 := &x.DeschedulingInterval
-		yym26 := z.DecBinary()
+		yym28 := z.DecBinary()
-		_ = yym26
+		_ = yym28
 		if false {
-		} else if yyxt26 := z.Extension(z.I2Rtid(yyv25)); yyxt26 != nil {
+		} else if yyxt28 := z.Extension(z.I2Rtid(yyv27)); yyxt28 != nil {
-			z.DecExtension(yyv25, yyxt26)
+			z.DecExtension(yyv27, yyxt28)
 		} else {
-			*((*int64)(yyv25)) = int64(r.DecodeInt(64))
+			*((*int64)(yyv27)) = int64(r.DecodeInt(64))
 		}
 	}
-	yyj20++
+	yyj22++
-	if yyhl20 {
+	if yyhl22 {
-		yyb20 = yyj20 > l
+		yyb22 = yyj22 > l
 	} else {
-		yyb20 = r.CheckBreak()
+		yyb22 = r.CheckBreak()
 	}
-	if yyb20 {
+	if yyb22 {
 		r.ReadArrayEnd()
 		return
 	}
@@ -551,29 +589,7 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 	if r.TryDecodeAsNil() {
 		x.KubeconfigFile = ""
 	} else {
-		yyv27 := &x.KubeconfigFile
+		yyv29 := &x.KubeconfigFile
-		yym28 := z.DecBinary()
-		_ = yym28
-		if false {
-		} else {
-			*((*string)(yyv27)) = r.DecodeString()
-		}
-	}
-	yyj20++
-	if yyhl20 {
-		yyb20 = yyj20 > l
-	} else {
-		yyb20 = r.CheckBreak()
-	}
-	if yyb20 {
-		r.ReadArrayEnd()
-		return
-	}
-	r.ReadArrayElem()
-	if r.TryDecodeAsNil() {
-		x.PolicyConfigFile = ""
-	} else {
-		yyv29 := &x.PolicyConfigFile
 		yym30 := z.DecBinary()
 		_ = yym30
 		if false {
@@ -581,13 +597,35 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 			*((*string)(yyv29)) = r.DecodeString()
 		}
 	}
-	yyj20++
+	yyj22++
-	if yyhl20 {
+	if yyhl22 {
-		yyb20 = yyj20 > l
+		yyb22 = yyj22 > l
 	} else {
-		yyb20 = r.CheckBreak()
+		yyb22 = r.CheckBreak()
 	}
-	if yyb20 {
+	if yyb22 {
+		r.ReadArrayEnd()
+		return
+	}
+	r.ReadArrayElem()
+	if r.TryDecodeAsNil() {
+		x.PolicyConfigFile = ""
+	} else {
+		yyv31 := &x.PolicyConfigFile
+		yym32 := z.DecBinary()
+		_ = yym32
+		if false {
+		} else {
+			*((*string)(yyv31)) = r.DecodeString()
+		}
+	}
+	yyj22++
+	if yyhl22 {
+		yyb22 = yyj22 > l
+	} else {
+		yyb22 = r.CheckBreak()
+	}
+	if yyb22 {
 		r.ReadArrayEnd()
 		return
 	}
@@ -595,21 +633,21 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 	if r.TryDecodeAsNil() {
 		x.DryRun = false
 	} else {
-		yyv31 := &x.DryRun
+		yyv33 := &x.DryRun
-		yym32 := z.DecBinary()
+		yym34 := z.DecBinary()
-		_ = yym32
+		_ = yym34
 		if false {
 		} else {
-			*((*bool)(yyv31)) = r.DecodeBool()
+			*((*bool)(yyv33)) = r.DecodeBool()
 		}
 	}
-	yyj20++
+	yyj22++
-	if yyhl20 {
+	if yyhl22 {
-		yyb20 = yyj20 > l
+		yyb22 = yyj22 > l
 	} else {
-		yyb20 = r.CheckBreak()
+		yyb22 = r.CheckBreak()
 	}
-	if yyb20 {
+	if yyb22 {
 		r.ReadArrayEnd()
 		return
 	}
@@ -617,21 +655,21 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 	if r.TryDecodeAsNil() {
 		x.NodeSelector = ""
 	} else {
-		yyv33 := &x.NodeSelector
+		yyv35 := &x.NodeSelector
-		yym34 := z.DecBinary()
+		yym36 := z.DecBinary()
-		_ = yym34
+		_ = yym36
 		if false {
 		} else {
-			*((*string)(yyv33)) = r.DecodeString()
+			*((*string)(yyv35)) = r.DecodeString()
 		}
 	}
-	yyj20++
+	yyj22++
-	if yyhl20 {
+	if yyhl22 {
-		yyb20 = yyj20 > l
+		yyb22 = yyj22 > l
 	} else {
-		yyb20 = r.CheckBreak()
+		yyb22 = r.CheckBreak()
 	}
-	if yyb20 {
+	if yyb22 {
 		r.ReadArrayEnd()
 		return
 	}
@@ -639,26 +677,48 @@ func (x *DeschedulerConfiguration) codecDecodeSelfFromArray(l int, d *codec1978.
 	if r.TryDecodeAsNil() {
 		x.MaxNoOfPodsToEvictPerNode = 0
 	} else {
-		yyv35 := &x.MaxNoOfPodsToEvictPerNode
+		yyv37 := &x.MaxNoOfPodsToEvictPerNode
-		yym36 := z.DecBinary()
+		yym38 := z.DecBinary()
-		_ = yym36
+		_ = yym38
 		if false {
 		} else {
-			*((*int)(yyv35)) = int(r.DecodeInt(codecSelferBitsize1234))
+			*((*int)(yyv37)) = int(r.DecodeInt(codecSelferBitsize1234))
+		}
+	}
+	yyj22++
+	if yyhl22 {
+		yyb22 = yyj22 > l
+	} else {
+		yyb22 = r.CheckBreak()
+	}
+	if yyb22 {
+		r.ReadArrayEnd()
+		return
+	}
+	r.ReadArrayElem()
+	if r.TryDecodeAsNil() {
+		x.EvictLocalStoragePods = false
+	} else {
+		yyv39 := &x.EvictLocalStoragePods
+		yym40 := z.DecBinary()
+		_ = yym40
+		if false {
+		} else {
+			*((*bool)(yyv39)) = r.DecodeBool()
 		}
 	}
 	for {
-		yyj20++
+		yyj22++
-		if yyhl20 {
+		if yyhl22 {
-			yyb20 = yyj20 > l
+			yyb22 = yyj22 > l
 		} else {
-			yyb20 = r.CheckBreak()
+			yyb22 = r.CheckBreak()
 		}
-		if yyb20 {
+		if yyb22 {
 			break
 		}
 		r.ReadArrayElem()
-		z.DecStructFieldNotFound(yyj20-1, "")
+		z.DecStructFieldNotFound(yyj22-1, "")
 	}
 	r.ReadArrayEnd()
 }

pkg/apis/componentconfig/v1alpha1/types.go

@@ -45,4 +45,7 @@ type DeschedulerConfiguration struct {
 
 	// MaxNoOfPodsToEvictPerNode restricts maximum of pods to be evicted per node.
 	MaxNoOfPodsToEvictPerNode int `json:"maxNoOfPodsToEvictPerNode,omitempty"`
+
+	// EvictLocalStoragePods allows pods using local storage to be evicted.
+	EvictLocalStoragePods bool `json:"evictLocalStoragePods,omitempty"`
 }

pkg/apis/componentconfig/v1alpha1/zz_generated.conversion.go

@@ -48,6 +48,7 @@ func autoConvert_v1alpha1_DeschedulerConfiguration_To_componentconfig_Deschedule
 	out.DryRun = in.DryRun
 	out.NodeSelector = in.NodeSelector
 	out.MaxNoOfPodsToEvictPerNode = in.MaxNoOfPodsToEvictPerNode
+	out.EvictLocalStoragePods = in.EvictLocalStoragePods
 	return nil
 }
 
@@ -63,6 +64,7 @@ func autoConvert_componentconfig_DeschedulerConfiguration_To_v1alpha1_Deschedule
 	out.DryRun = in.DryRun
 	out.NodeSelector = in.NodeSelector
 	out.MaxNoOfPodsToEvictPerNode = in.MaxNoOfPodsToEvictPerNode
+	out.EvictLocalStoragePods = in.EvictLocalStoragePods
 	return nil
 }
 

pkg/descheduler/pod/pods.go

@@ -52,23 +52,23 @@ func IsLatencySensitivePod(pod *v1.Pod) bool {
 }
 
 // IsEvictable checks if a pod is evictable or not.
-func IsEvictable(pod *v1.Pod) bool {
+func IsEvictable(pod *v1.Pod, evictLocalStoragePods bool) bool {
 	ownerRefList := OwnerRef(pod)
-	if IsMirrorPod(pod) || IsPodWithLocalStorage(pod) || len(ownerRefList) == 0 || IsDaemonsetPod(ownerRefList) || IsCriticalPod(pod) {
+	if IsMirrorPod(pod) || (!evictLocalStoragePods && IsPodWithLocalStorage(pod)) || len(ownerRefList) == 0 || IsDaemonsetPod(ownerRefList) || IsCriticalPod(pod) {
 		return false
 	}
 	return true
 }
 
 // ListEvictablePodsOnNode returns the list of evictable pods on node.
-func ListEvictablePodsOnNode(client clientset.Interface, node *v1.Node) ([]*v1.Pod, error) {
+func ListEvictablePodsOnNode(client clientset.Interface, node *v1.Node, evictLocalStoragePods bool) ([]*v1.Pod, error) {
 	pods, err := ListPodsOnANode(client, node)
 	if err != nil {
 		return []*v1.Pod{}, err
 	}
 	evictablePods := make([]*v1.Pod, 0)
 	for _, pod := range pods {
-		if !IsEvictable(pod) {
+		if !IsEvictable(pod, evictLocalStoragePods) {
 			continue
 		} else {
 			evictablePods = append(evictablePods, pod)

pkg/descheduler/strategies/duplicates.go

@@ -40,15 +40,15 @@ func RemoveDuplicatePods(ds *options.DeschedulerServer, strategy api.Descheduler
 	if !strategy.Enabled {
 		return
 	}
-	deleteDuplicatePods(ds.Client, policyGroupVersion, nodes, ds.DryRun, nodepodCount, ds.MaxNoOfPodsToEvictPerNode)
+	deleteDuplicatePods(ds.Client, policyGroupVersion, nodes, ds.DryRun, nodepodCount, ds.MaxNoOfPodsToEvictPerNode, ds.EvictLocalStoragePods)
 }
 
 // deleteDuplicatePods evicts the pod from node and returns the count of evicted pods.
-func deleteDuplicatePods(client clientset.Interface, policyGroupVersion string, nodes []*v1.Node, dryRun bool, nodepodCount nodePodEvictedCount, maxPodsToEvict int) int {
+func deleteDuplicatePods(client clientset.Interface, policyGroupVersion string, nodes []*v1.Node, dryRun bool, nodepodCount nodePodEvictedCount, maxPodsToEvict int, evictLocalStoragePods bool) int {
 	podsEvicted := 0
 	for _, node := range nodes {
 		glog.V(1).Infof("Processing node: %#v", node.Name)
-		dpm := ListDuplicatePodsOnANode(client, node)
+		dpm := ListDuplicatePodsOnANode(client, node, evictLocalStoragePods)
 		for creator, pods := range dpm {
 			if len(pods) > 1 {
 				glog.V(1).Infof("%#v", creator)
@@ -73,8 +73,8 @@ func deleteDuplicatePods(client clientset.Interface, policyGroupVersion string,
 }
 
 // ListDuplicatePodsOnANode lists duplicate pods on a given node.
-func ListDuplicatePodsOnANode(client clientset.Interface, node *v1.Node) DuplicatePodsMap {
+func ListDuplicatePodsOnANode(client clientset.Interface, node *v1.Node, evictLocalStoragePods bool) DuplicatePodsMap {
-	pods, err := podutil.ListEvictablePodsOnNode(client, node)
+	pods, err := podutil.ListEvictablePodsOnNode(client, node, evictLocalStoragePods)
 	if err != nil {
 		return nil
 	}

pkg/descheduler/strategies/duplicates_test.go

@@ -77,7 +77,7 @@ func TestFindDuplicatePods(t *testing.T) {
 	})
 	npe := nodePodEvictedCount{}
 	npe[node] = 0
-	podsEvicted := deleteDuplicatePods(fakeClient, "v1", []*v1.Node{node}, false, npe, 2)
+	podsEvicted := deleteDuplicatePods(fakeClient, "v1", []*v1.Node{node}, false, npe, 2, false)
 	if podsEvicted != expectedEvictedPodCount {
 		t.Errorf("Unexpected no of pods evicted")
 	}

pkg/descheduler/strategies/lownodeutilization.go

@@ -41,6 +41,7 @@ type NodeUsageMap struct {
 	bPods            []*v1.Pod
 	gPods            []*v1.Pod
 }
+
 type NodePodsMap map[*v1.Node][]*v1.Pod
 
 func LowNodeUtilization(ds *options.DeschedulerServer, strategy api.DeschedulerStrategy, evictionPolicyGroupVersion string, nodes []*v1.Node, nodepodCount nodePodEvictedCount) {
@@ -59,8 +60,8 @@ func LowNodeUtilization(ds *options.DeschedulerServer, strategy api.DeschedulerS
 		return
 	}
 
-	npm := CreateNodePodsMap(ds.Client, nodes)
+	npm := createNodePodsMap(ds.Client, nodes)
-	lowNodes, targetNodes := classifyNodes(npm, thresholds, targetThresholds)
+	lowNodes, targetNodes := classifyNodes(npm, thresholds, targetThresholds, ds.EvictLocalStoragePods)
 
 	glog.V(1).Infof("Criteria for a node under utilization: CPU: %v, Mem: %v, Pods: %v",
 		thresholds[v1.ResourceCPU], thresholds[v1.ResourceMemory], thresholds[v1.ResourcePods])
@@ -130,10 +131,10 @@ func validateTargetThresholds(targetThresholds api.ResourceThresholds) bool {
 
 // classifyNodes classifies the nodes into low-utilization or high-utilization nodes. If a node lies between
 // low and high thresholds, it is simply ignored.
-func classifyNodes(npm NodePodsMap, thresholds api.ResourceThresholds, targetThresholds api.ResourceThresholds) ([]NodeUsageMap, []NodeUsageMap) {
+func classifyNodes(npm NodePodsMap, thresholds api.ResourceThresholds, targetThresholds api.ResourceThresholds, evictLocalStoragePods bool) ([]NodeUsageMap, []NodeUsageMap) {
 	lowNodes, targetNodes := []NodeUsageMap{}, []NodeUsageMap{}
 	for node, pods := range npm {
-		usage, allPods, nonRemovablePods, bePods, bPods, gPods := NodeUtilization(node, pods)
+		usage, allPods, nonRemovablePods, bePods, bPods, gPods := NodeUtilization(node, pods, evictLocalStoragePods)
 		nuMap := NodeUsageMap{node, usage, allPods, nonRemovablePods, bePods, bPods, gPods}
 
 		// Check if node is underutilized and if we can schedule pods on it.
@@ -151,6 +152,9 @@ func classifyNodes(npm NodePodsMap, thresholds api.ResourceThresholds, targetThr
 	return lowNodes, targetNodes
 }
 
+// evictPodsFromTargetNodes evicts pods based on priority, if all the pods on the node have priority, if not
+// evicts them based on QoS as fallback option.
+// TODO: @ravig Break this function into smaller functions.
 func evictPodsFromTargetNodes(client clientset.Interface, evictionPolicyGroupVersion string, targetNodes, lowNodes []NodeUsageMap, targetThresholds api.ResourceThresholds, dryRun bool, maxPodsToEvict int, nodepodCount nodePodEvictedCount) int {
 	podsEvicted := 0
 
@@ -191,12 +195,27 @@ func evictPodsFromTargetNodes(client clientset.Interface, evictionPolicyGroupVer
 		glog.V(3).Infof("evicting pods from node %#v with usage: %#v", node.node.Name, node.usage)
 		currentPodsEvicted := nodepodCount[node.node]
 
-		// evict best effort pods
+		// Check if one pod has priority, if yes, assume that all pods have priority and evict pods based on priority.
-		evictPods(node.bePods, client, evictionPolicyGroupVersion, targetThresholds, nodeCapacity, node.usage, &totalPods, &totalCpu, &totalMem, &currentPodsEvicted, dryRun, maxPodsToEvict)
+		if node.allPods[0].Spec.Priority != nil {
-		// evict burstable pods
+			glog.V(1).Infof("All pods have priority associated with them. Evicting pods based on priority")
-		evictPods(node.bPods, client, evictionPolicyGroupVersion, targetThresholds, nodeCapacity, node.usage, &totalPods, &totalCpu, &totalMem, &currentPodsEvicted, dryRun, maxPodsToEvict)
+			evictablePods := make([]*v1.Pod, 0)
-		// evict guaranteed pods
+			evictablePods = append(append(node.bPods, node.bePods...), node.gPods...)
-		evictPods(node.gPods, client, evictionPolicyGroupVersion, targetThresholds, nodeCapacity, node.usage, &totalPods, &totalCpu, &totalMem, &currentPodsEvicted, dryRun, maxPodsToEvict)
+
+			// sort the evictable Pods based on priority. This also sorts them based on QoS. If there are multiple pods with same priority, they are sorted based on QoS tiers.
+			sortPodsBasedOnPriority(evictablePods)
+			evictPods(evictablePods, client, evictionPolicyGroupVersion, targetThresholds, nodeCapacity, node.usage, &totalPods, &totalCpu, &totalMem, &currentPodsEvicted, dryRun, maxPodsToEvict)
+		} else {
+			// TODO: Remove this when we support only priority.
+			//  Falling back to evicting pods based on priority.
+			glog.V(1).Infof("Evicting pods based on QoS")
+			glog.V(1).Infof("There are %v non-evictable pods on the node", len(node.nonRemovablePods))
+			// evict best effort pods
+			evictPods(node.bePods, client, evictionPolicyGroupVersion, targetThresholds, nodeCapacity, node.usage, &totalPods, &totalCpu, &totalMem, &currentPodsEvicted, dryRun, maxPodsToEvict)
+			// evict burstable pods
+			evictPods(node.bPods, client, evictionPolicyGroupVersion, targetThresholds, nodeCapacity, node.usage, &totalPods, &totalCpu, &totalMem, &currentPodsEvicted, dryRun, maxPodsToEvict)
+			// evict guaranteed pods
+			evictPods(node.gPods, client, evictionPolicyGroupVersion, targetThresholds, nodeCapacity, node.usage, &totalPods, &totalCpu, &totalMem, &currentPodsEvicted, dryRun, maxPodsToEvict)
+		}
 		nodepodCount[node.node] = currentPodsEvicted
 		podsEvicted = podsEvicted + nodepodCount[node.node]
 		glog.V(1).Infof("%v pods evicted from node %#v with usage %v", nodepodCount[node.node], node.node.Name, node.usage)
@@ -269,7 +288,30 @@ func SortNodesByUsage(nodes []NodeUsageMap) {
 	})
 }
 
-func CreateNodePodsMap(client clientset.Interface, nodes []*v1.Node) NodePodsMap {
+// sortPodsBasedOnPriority sorts pods based on priority and if their priorities are equal, they are sorted based on QoS tiers.
+func sortPodsBasedOnPriority(evictablePods []*v1.Pod) {
+	sort.Slice(evictablePods, func(i, j int) bool {
+		if evictablePods[i].Spec.Priority == nil && evictablePods[j].Spec.Priority != nil {
+			return true
+		}
+		if evictablePods[j].Spec.Priority == nil && evictablePods[i].Spec.Priority != nil {
+			return false
+		}
+		if (evictablePods[j].Spec.Priority == nil && evictablePods[i].Spec.Priority == nil) || (*evictablePods[i].Spec.Priority == *evictablePods[j].Spec.Priority) {
+			if podutil.IsBestEffortPod(evictablePods[i]) {
+				return true
+			}
+			if podutil.IsBurstablePod(evictablePods[i]) && podutil.IsGuaranteedPod(evictablePods[j]) {
+				return true
+			}
+			return false
+		}
+		return *evictablePods[i].Spec.Priority < *evictablePods[j].Spec.Priority
+	})
+}
+
+// createNodePodsMap returns nodepodsmap with evictable pods on node.
+func createNodePodsMap(client clientset.Interface, nodes []*v1.Node) NodePodsMap {
 	npm := NodePodsMap{}
 	for _, node := range nodes {
 		pods, err := podutil.ListPodsOnANode(client, node)
@@ -308,7 +350,8 @@ func IsNodeWithLowUtilization(nodeThresholds api.ResourceThresholds, thresholds
 	return true
 }
 
-func NodeUtilization(node *v1.Node, pods []*v1.Pod) (api.ResourceThresholds, []*v1.Pod, []*v1.Pod, []*v1.Pod, []*v1.Pod, []*v1.Pod) {
+// Nodeutilization returns the current usage of node.
+func NodeUtilization(node *v1.Node, pods []*v1.Pod, evictLocalStoragePods bool) (api.ResourceThresholds, []*v1.Pod, []*v1.Pod, []*v1.Pod, []*v1.Pod, []*v1.Pod) {
 	bePods := []*v1.Pod{}
 	nonRemovablePods := []*v1.Pod{}
 	bPods := []*v1.Pod{}
@@ -316,7 +359,7 @@ func NodeUtilization(node *v1.Node, pods []*v1.Pod) (api.ResourceThresholds, []*
 	totalReqs := map[v1.ResourceName]resource.Quantity{}
 	for _, pod := range pods {
 		// We need to compute the usage of nonRemovablePods unless it is a best effort pod. So, cannot use podutil.ListEvictablePodsOnNode
-		if !podutil.IsEvictable(pod) {
+		if !podutil.IsEvictable(pod, evictLocalStoragePods) {
 			nonRemovablePods = append(nonRemovablePods, pod)
 			if podutil.IsBestEffortPod(pod) {
 				continue

pkg/descheduler/strategies/lownodeutilization_test.go

@@ -28,10 +28,11 @@ import (
 	"k8s.io/apimachinery/pkg/runtime"
 	"k8s.io/client-go/kubernetes/fake"
 	core "k8s.io/client-go/testing"
+	"reflect"
 )
 
 // TODO: Make this table driven.
-func TestLowNodeUtilization(t *testing.T) {
+func TestLowNodeUtilizationWithoutPriority(t *testing.T) {
 	var thresholds = make(api.ResourceThresholds)
 	var targetThresholds = make(api.ResourceThresholds)
 	thresholds[v1.ResourceCPU] = 30
@@ -110,8 +111,8 @@ func TestLowNodeUtilization(t *testing.T) {
 		return true, nil, fmt.Errorf("Wrong node: %v", getAction.GetName())
 	})
 	expectedPodsEvicted := 3
-	npm := CreateNodePodsMap(fakeClient, []*v1.Node{n1, n2, n3})
+	npm := createNodePodsMap(fakeClient, []*v1.Node{n1, n2, n3})
-	lowNodes, targetNodes := classifyNodes(npm, thresholds, targetThresholds)
+	lowNodes, targetNodes := classifyNodes(npm, thresholds, targetThresholds, false)
 	if len(lowNodes) != 1 {
 		t.Errorf("After ignoring unschedulable nodes, expected only one node to be under utilized.")
 	}
@@ -126,6 +127,151 @@ func TestLowNodeUtilization(t *testing.T) {
 
 }
 
+// TODO: Make this table driven.
+func TestLowNodeUtilizationWithPriorities(t *testing.T) {
+	var thresholds = make(api.ResourceThresholds)
+	var targetThresholds = make(api.ResourceThresholds)
+	thresholds[v1.ResourceCPU] = 30
+	thresholds[v1.ResourcePods] = 30
+	targetThresholds[v1.ResourceCPU] = 50
+	targetThresholds[v1.ResourcePods] = 50
+	lowPriority := int32(0)
+	highPriority := int32(10000)
+	n1 := test.BuildTestNode("n1", 4000, 3000, 9)
+	n2 := test.BuildTestNode("n2", 4000, 3000, 10)
+	n3 := test.BuildTestNode("n3", 4000, 3000, 10)
+	// Making n3 node unschedulable so that it won't counted in lowUtilized nodes list.
+	n3.Spec.Unschedulable = true
+	p1 := test.BuildTestPod("p1", 400, 0, n1.Name)
+	p1.Spec.Priority = &highPriority
+	p2 := test.BuildTestPod("p2", 400, 0, n1.Name)
+	p2.Spec.Priority = &highPriority
+	p3 := test.BuildTestPod("p3", 400, 0, n1.Name)
+	p3.Spec.Priority = &highPriority
+	p4 := test.BuildTestPod("p4", 400, 0, n1.Name)
+	p4.Spec.Priority = &highPriority
+	p5 := test.BuildTestPod("p5", 400, 0, n1.Name)
+	p5.Spec.Priority = &lowPriority
+
+	// These won't be evicted.
+	p6 := test.BuildTestPod("p6", 400, 0, n1.Name)
+	p6.Spec.Priority = &highPriority
+	p7 := test.BuildTestPod("p7", 400, 0, n1.Name)
+	p7.Spec.Priority = &lowPriority
+	p8 := test.BuildTestPod("p8", 400, 0, n1.Name)
+	p8.Spec.Priority = &lowPriority
+
+	p1.ObjectMeta.OwnerReferences = test.GetReplicaSetOwnerRefList()
+	p2.ObjectMeta.OwnerReferences = test.GetReplicaSetOwnerRefList()
+	p3.ObjectMeta.OwnerReferences = test.GetReplicaSetOwnerRefList()
+	p4.ObjectMeta.OwnerReferences = test.GetReplicaSetOwnerRefList()
+	p5.ObjectMeta.OwnerReferences = test.GetReplicaSetOwnerRefList()
+	// The following 4 pods won't get evicted.
+	// A daemonset.
+	p6.ObjectMeta.OwnerReferences = test.GetDaemonSetOwnerRefList()
+	// A pod with local storage.
+	p7.ObjectMeta.OwnerReferences = test.GetNormalPodOwnerRefList()
+	p7.Spec.Volumes = []v1.Volume{
+		{
+			Name: "sample",
+			VolumeSource: v1.VolumeSource{
+				HostPath: &v1.HostPathVolumeSource{Path: "somePath"},
+				EmptyDir: &v1.EmptyDirVolumeSource{
+					SizeLimit: resource.NewQuantity(int64(10), resource.BinarySI)},
+			},
+		},
+	}
+	// A Mirror Pod.
+	p7.Annotations = test.GetMirrorPodAnnotation()
+	// A Critical Pod.
+	p8.Namespace = "kube-system"
+	p8.Annotations = test.GetCriticalPodAnnotation()
+	p9 := test.BuildTestPod("p9", 400, 0, n1.Name)
+	p9.ObjectMeta.OwnerReferences = test.GetReplicaSetOwnerRefList()
+	fakeClient := &fake.Clientset{}
+	fakeClient.Fake.AddReactor("list", "pods", func(action core.Action) (bool, runtime.Object, error) {
+		list := action.(core.ListAction)
+		fieldString := list.GetListRestrictions().Fields.String()
+		if strings.Contains(fieldString, "n1") {
+			return true, &v1.PodList{Items: []v1.Pod{*p1, *p2, *p3, *p4, *p5, *p6, *p7, *p8}}, nil
+		}
+		if strings.Contains(fieldString, "n2") {
+			return true, &v1.PodList{Items: []v1.Pod{*p9}}, nil
+		}
+		if strings.Contains(fieldString, "n3") {
+			return true, &v1.PodList{Items: []v1.Pod{}}, nil
+		}
+		return true, nil, fmt.Errorf("Failed to list: %v", list)
+	})
+	fakeClient.Fake.AddReactor("get", "nodes", func(action core.Action) (bool, runtime.Object, error) {
+		getAction := action.(core.GetAction)
+		switch getAction.GetName() {
+		case n1.Name:
+			return true, n1, nil
+		case n2.Name:
+			return true, n2, nil
+		case n3.Name:
+			return true, n3, nil
+		}
+		return true, nil, fmt.Errorf("Wrong node: %v", getAction.GetName())
+	})
+	expectedPodsEvicted := 3
+	npm := createNodePodsMap(fakeClient, []*v1.Node{n1, n2, n3})
+	lowNodes, targetNodes := classifyNodes(npm, thresholds, targetThresholds, false)
+	if len(lowNodes) != 1 {
+		t.Errorf("After ignoring unschedulable nodes, expected only one node to be under utilized.")
+	}
+	npe := nodePodEvictedCount{}
+	npe[n1] = 0
+	npe[n2] = 0
+	npe[n3] = 0
+	podsEvicted := evictPodsFromTargetNodes(fakeClient, "v1", targetNodes, lowNodes, targetThresholds, false, 3, npe)
+	if expectedPodsEvicted != podsEvicted {
+		t.Errorf("Expected %#v pods to be evicted but %#v got evicted", expectedPodsEvicted, podsEvicted)
+	}
+
+}
+
+func TestSortPodsByPriority(t *testing.T) {
+	n1 := test.BuildTestNode("n1", 4000, 3000, 9)
+	lowPriority := int32(0)
+	highPriority := int32(10000)
+	p1 := test.BuildTestPod("p1", 400, 0, n1.Name)
+	p1.Spec.Priority = &lowPriority
+
+	// BestEffort
+	p2 := test.BuildTestPod("p2", 400, 0, n1.Name)
+	p2.Spec.Priority = &highPriority
+
+	p2.Spec.Containers[0].Resources.Requests = nil
+	p2.Spec.Containers[0].Resources.Limits = nil
+
+	// Burstable
+	p3 := test.BuildTestPod("p3", 400, 0, n1.Name)
+	p3.Spec.Priority = &highPriority
+
+	// Guaranteed
+	p4 := test.BuildTestPod("p4", 400, 100, n1.Name)
+	p4.Spec.Priority = &highPriority
+	p4.Spec.Containers[0].Resources.Limits[v1.ResourceCPU] = *resource.NewMilliQuantity(400, resource.DecimalSI)
+	p4.Spec.Containers[0].Resources.Limits[v1.ResourceMemory] = *resource.NewQuantity(100, resource.DecimalSI)
+
+	// Best effort with nil priorities.
+	p5 := test.BuildTestPod("p5", 400, 100, n1.Name)
+	p5.Spec.Priority = nil
+	p6 := test.BuildTestPod("p6", 400, 100, n1.Name)
+	p6.Spec.Containers[0].Resources.Limits[v1.ResourceCPU] = *resource.NewMilliQuantity(400, resource.DecimalSI)
+	p6.Spec.Containers[0].Resources.Limits[v1.ResourceMemory] = *resource.NewQuantity(100, resource.DecimalSI)
+	p6.Spec.Priority = nil
+
+	podList := []*v1.Pod{p4, p3, p2, p1, p6, p5}
+
+	sortPodsBasedOnPriority(podList)
+	if !reflect.DeepEqual(podList[len(podList)-1], p4) {
+		t.Errorf("Expected last pod in sorted list to be %v which of highest priority and guaranteed but got %v", p4, podList[len(podList)-1])
+	}
+}
+
 func TestValidateThresholds(t *testing.T) {
 	tests := []struct {
 		name    string

pkg/descheduler/strategies/node_affinity.go

@@ -27,10 +27,10 @@ import (
 )
 
 func RemovePodsViolatingNodeAffinity(ds *options.DeschedulerServer, strategy api.DeschedulerStrategy, evictionPolicyGroupVersion string, nodes []*v1.Node, nodePodCount nodePodEvictedCount) {
-	removePodsViolatingNodeAffinityCount(ds, strategy, evictionPolicyGroupVersion, nodes, nodePodCount, ds.MaxNoOfPodsToEvictPerNode)
+	removePodsViolatingNodeAffinityCount(ds, strategy, evictionPolicyGroupVersion, nodes, nodePodCount, ds.MaxNoOfPodsToEvictPerNode, ds.EvictLocalStoragePods)
 }
 
-func removePodsViolatingNodeAffinityCount(ds *options.DeschedulerServer, strategy api.DeschedulerStrategy, evictionPolicyGroupVersion string, nodes []*v1.Node, nodepodCount nodePodEvictedCount, maxPodsToEvict int) int {
+func removePodsViolatingNodeAffinityCount(ds *options.DeschedulerServer, strategy api.DeschedulerStrategy, evictionPolicyGroupVersion string, nodes []*v1.Node, nodepodCount nodePodEvictedCount, maxPodsToEvict int, evictLocalStoragePods bool) int {
 	evictedPodCount := 0
 	if !strategy.Enabled {
 		return evictedPodCount
@@ -44,7 +44,7 @@ func removePodsViolatingNodeAffinityCount(ds *options.DeschedulerServer, strateg
 			for _, node := range nodes {
 				glog.V(1).Infof("Processing node: %#v\n", node.Name)
 
-				pods, err := podutil.ListEvictablePodsOnNode(ds.Client, node)
+				pods, err := podutil.ListEvictablePodsOnNode(ds.Client, node, evictLocalStoragePods)
 				if err != nil {
 					glog.Errorf("failed to get pods from %v: %v", node.Name, err)
 				}

pkg/descheduler/strategies/node_affinity_test.go

@@ -176,7 +176,7 @@ func TestRemovePodsViolatingNodeAffinity(t *testing.T) {
 			Client: fakeClient,
 		}
 
-		actualEvictedPodCount := removePodsViolatingNodeAffinityCount(&ds, tc.strategy, "v1", tc.nodes, tc.npe, tc.maxPodsToEvict)
+		actualEvictedPodCount := removePodsViolatingNodeAffinityCount(&ds, tc.strategy, "v1", tc.nodes, tc.npe, tc.maxPodsToEvict, false)
 		if actualEvictedPodCount != tc.expectedEvictedPodCount {
 			t.Errorf("Test %#v failed, expected %v pod evictions, but got %v pod evictions\n", tc.description, tc.expectedEvictedPodCount, actualEvictedPodCount)
 		}

pkg/descheduler/strategies/pod_antiaffinity.go

@@ -35,15 +35,15 @@ func RemovePodsViolatingInterPodAntiAffinity(ds *options.DeschedulerServer, stra
 	if !strategy.Enabled {
 		return
 	}
-	removePodsWithAffinityRules(ds.Client, policyGroupVersion, nodes, ds.DryRun, nodePodCount, ds.MaxNoOfPodsToEvictPerNode)
+	removePodsWithAffinityRules(ds.Client, policyGroupVersion, nodes, ds.DryRun, nodePodCount, ds.MaxNoOfPodsToEvictPerNode, ds.EvictLocalStoragePods)
 }
 
 // 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, nodePodCount nodePodEvictedCount, maxPodsToEvict int) int {
+func removePodsWithAffinityRules(client clientset.Interface, policyGroupVersion string, nodes []*v1.Node, dryRun bool, nodePodCount nodePodEvictedCount, maxPodsToEvict int, evictLocalStoragePods bool) int {
 	podsEvicted := 0
 	for _, node := range nodes {
 		glog.V(1).Infof("Processing node: %#v\n", node.Name)
-		pods, err := podutil.ListEvictablePodsOnNode(client, node)
+		pods, err := podutil.ListEvictablePodsOnNode(client, node, evictLocalStoragePods)
 		if err != nil {
 			return 0
 		}

pkg/descheduler/strategies/pod_antiaffinity_test.go

@@ -55,13 +55,13 @@ func TestPodAntiAffinity(t *testing.T) {
 	npe := nodePodEvictedCount{}
 	npe[node] = 0
 	expectedEvictedPodCount := 3
-	podsEvicted := removePodsWithAffinityRules(fakeClient, "v1", []*v1.Node{node}, false, npe, 0)
+	podsEvicted := removePodsWithAffinityRules(fakeClient, "v1", []*v1.Node{node}, false, npe, 0, false)
 	if podsEvicted != expectedEvictedPodCount {
 		t.Errorf("Unexpected no of pods evicted: pods evicted: %d, expected: %d", podsEvicted, expectedEvictedPodCount)
 	}
 	npe[node] = 0
 	expectedEvictedPodCount = 1
-	podsEvicted = removePodsWithAffinityRules(fakeClient, "v1", []*v1.Node{node}, false, npe, 1)
+	podsEvicted = removePodsWithAffinityRules(fakeClient, "v1", []*v1.Node{node}, false, npe, 1, false)
 	if podsEvicted != expectedEvictedPodCount {
 		t.Errorf("Unexpected no of pods evicted: pods evicted: %d, expected: %d", podsEvicted, expectedEvictedPodCount)
 	}

test/test_utils.go

@@ -37,6 +37,7 @@ func BuildTestPod(name string, cpu int64, memory int64, nodeName string) *v1.Pod
 				{
 					Resources: v1.ResourceRequirements{
 						Requests: v1.ResourceList{},
+						Limits:   v1.ResourceList{},
 					},
 				},
 			},