Working nodeFit feature

README.md

@@ -157,6 +157,7 @@ should include `ReplicaSet` to have pods created by Deployments excluded.
 |`namespaces`|(see [namespace filtering](#namespace-filtering))|
 |`thresholdPriority`|int (see [priority filtering](#priority-filtering))|
 |`thresholdPriorityClassName`|string (see [priority filtering](#priority-filtering))|
+|`nodeFit`|string (see [node fit filtering](#node-fit-filtering))|
 
 **Example:**
 ```yaml
@@ -204,6 +205,7 @@ strategy evicts pods from `overutilized nodes` (those with usage above `targetTh
 |`numberOfNodes`|int|
 |`thresholdPriority`|int (see [priority filtering](#priority-filtering))|
 |`thresholdPriorityClassName`|string (see [priority filtering](#priority-filtering))|
+|`nodeFit`|string (see [node fit filtering](#node-fit-filtering))|
 
 **Example:**
 
@@ -253,6 +255,7 @@ node.
 |`thresholdPriorityClassName`|string (see [priority filtering](#priority-filtering))|
 |`namespaces`|(see [namespace filtering](#namespace-filtering))|
 |`labelSelector`|(see [label filtering](#label-filtering))|
+|`nodeFit`|string (see [node fit filtering](#node-fit-filtering))|
 
 **Example:**
 
@@ -291,6 +294,7 @@ podA gets evicted from nodeA.
 |`thresholdPriorityClassName`|string (see [priority filtering](#priority-filtering))|
 |`namespaces`|(see [namespace filtering](#namespace-filtering))|
 |`labelSelector`|(see [label filtering](#label-filtering))|
+|`nodeFit`|string (see [node fit filtering](#node-fit-filtering))|
 
 **Example:**
 
@@ -320,6 +324,7 @@ and will be evicted.
 |`thresholdPriorityClassName`|string (see [priority filtering](#priority-filtering))|
 |`namespaces`|(see [namespace filtering](#namespace-filtering))|
 |`labelSelector`|(see [label filtering](#label-filtering))|
+|`nodeFit`|string (see [node fit filtering](#node-fit-filtering))|
 
 **Example:**
 
@@ -348,6 +353,7 @@ include soft constraints.
 |`thresholdPriority`|int (see [priority filtering](#priority-filtering))|
 |`thresholdPriorityClassName`|string (see [priority filtering](#priority-filtering))|
 |`namespaces`|(see [namespace filtering](#namespace-filtering))|
+|`nodeFit`|string (see [node fit filtering](#node-fit-filtering))|
 
 **Example:**
 
@@ -379,6 +385,7 @@ which determines whether init container restarts should be factored into that ca
 |`thresholdPriority`|int (see [priority filtering](#priority-filtering))|
 |`thresholdPriorityClassName`|string (see [priority filtering](#priority-filtering))|
 |`namespaces`|(see [namespace filtering](#namespace-filtering))|
+|`nodeFit`|string (see [node fit filtering](#node-fit-filtering))|
 
 **Example:**
 
@@ -411,6 +418,7 @@ to `Running` and `Pending`.
 |`thresholdPriorityClassName`|string (see [priority filtering](#priority-filtering))|
 |`namespaces`|(see [namespace filtering](#namespace-filtering))|
 |`labelSelector`|(see [label filtering](#label-filtering))|
+|`nodeFit`|string (see [node fit filtering](#node-fit-filtering))|
 
 **Example:**
 
@@ -551,6 +559,30 @@ strategies:
           - {key: environment, operator: NotIn, values: [dev]}
 ```
 
+
+### Node Fit filtering
+
+All strategies are able to configure a `nodeFit` boolean parameter. If set to `true` the descheduler will consider whether or not the pods that meet eviction criteria will fit on other nodes before evicting them. If a pod cannot be rescheduled to another node, it will not be evicted. Currently the following criteria are considered when setting `nodeFit` to `true`:
+- A `nodeSelector` on the pod
+- Any `Tolerations` on the pod and any `Taints` on the other nodes
+- `nodeAffinity` on the pod
+- Whether any of the other nodes are marked as `unschedulable`
+
+E.g.
+
+```yaml
+apiVersion: "descheduler/v1alpha1"
+kind: "DeschedulerPolicy"
+strategies:
+  "PodLifeTime":
+     enabled: true
+     params:
+        podLifeTime:
+          maxPodLifeTimeSeconds: 86400
+        nodeFit: true
+```
+
+
 ## Pod Evictions
 
 When the descheduler decides to evict pods from a node, it employs the following general mechanism:

pkg/api/types.go

@@ -80,6 +80,7 @@ type StrategyParameters struct {
 	ThresholdPriority                 *int32
 	ThresholdPriorityClassName        string
 	LabelSelector                     *metav1.LabelSelector
+	NodeFit                           bool
 }
 
 type Percentage float64

pkg/api/v1alpha1/types.go

@@ -78,6 +78,7 @@ type StrategyParameters struct {
 	ThresholdPriority                 *int32                             `json:"thresholdPriority"`
 	ThresholdPriorityClassName        string                             `json:"thresholdPriorityClassName"`
 	LabelSelector                     *metav1.LabelSelector              `json:"labelSelector"`
+	NodeFit                           bool                               `json:"nodeFit"`
 }
 
 type Percentage float64

pkg/descheduler/evictions/evictions.go

@@ -32,6 +32,7 @@ import (
 	"k8s.io/client-go/tools/record"
 	"k8s.io/klog/v2"
 	"sigs.k8s.io/descheduler/metrics"
+	nodeutil "sigs.k8s.io/descheduler/pkg/descheduler/node"
 	podutil "sigs.k8s.io/descheduler/pkg/descheduler/pod"
 	"sigs.k8s.io/descheduler/pkg/utils"
 
@@ -47,6 +48,7 @@ type nodePodEvictedCount map[*v1.Node]int
 
 type PodEvictor struct {
 	client                  clientset.Interface
+	nodes                   []*v1.Node
 	policyGroupVersion      string
 	dryRun                  bool
 	maxPodsToEvictPerNode   int
@@ -74,6 +76,7 @@ func NewPodEvictor(
 
 	return &PodEvictor{
 		client:                  client,
+		nodes:                   nodes,
 		policyGroupVersion:      policyGroupVersion,
 		dryRun:                  dryRun,
 		maxPodsToEvictPerNode:   maxPodsToEvictPerNode,
@@ -164,6 +167,7 @@ func evictPod(ctx context.Context, client clientset.Interface, pod *v1.Pod, poli
 
 type Options struct {
 	priority *int32
+	nodeFit  *bool
 }
 
 // WithPriorityThreshold sets a threshold for pod's priority class.
@@ -175,6 +179,17 @@ func WithPriorityThreshold(priority int32) func(opts *Options) {
 	}
 }
 
+// WithNodeFit sets whether or not to consider taints, node selectors,
+// and pod affinity when evicting. A pod who's tolerations, node selectors,
+// and affinity match a node other than the one it is currently running on
+// is evictable.
+func WithNodeFit(nodeFit bool) func(opts *Options) {
+	return func(opts *Options) {
+		var a bool = nodeFit
+		opts.nodeFit = &a
+	}
+}
+
 type constraint func(pod *v1.Pod) error
 
 type evictable struct {
@@ -225,6 +240,14 @@ func (pe *PodEvictor) Evictable(opts ...func(opts *Options)) *evictable {
 			return nil
 		})
 	}
+	if options.nodeFit != nil && *options.nodeFit {
+		ev.constraints = append(ev.constraints, func(pod *v1.Pod) error {
+			if !nodeutil.PodFitsAnyOtherNode(pod, pe.nodes) {
+				return fmt.Errorf("pod does not fit on any other node because of nodeSelector(s), Taint(s), or nodes marked as unschedulable")
+			}
+			return nil
+		})
+	}
 
 	return ev
 }