Why Do Kubernetes Pods Get Evicted? Understanding Preemption, QoS, and Node Pressure

What Does It Mean When a Kubernetes Pod Is Evicted?

Pods are terminated, usually because there are not enough resources. The eviction process is the termination of a pod assigned to a node.

Reasons for Pod Eviction: Preemption and Node Pressure

One common case is Preemption : when a new pod needs to be scheduled on a resource‑constrained node, the scheduler may evict lower‑priority pods to free resources.

Kubernetes continuously checks node resources and evicts pods when necessary; this is called node‑pressure eviction .

Preemptive Eviction

If no suitable node has enough resources for a new pod, the kube-scheduler evicts lower‑priority pods to make room.

Pod Scheduling

Scheduling assigns pods to nodes. The kube-scheduler runs in the control plane, keeping pods in Pending state until a matching node is found. The scheduling process consists of:

Filtering

Scoring

Filtering

During filtering, the scheduler selects all nodes where the pod could be placed, considering taints, tolerations, and other constraints.

Scoring

In the scoring step, each candidate node receives a score; the highest‑scoring node is chosen. Ties are broken randomly.

If no node can accommodate the pod, the scheduler initiates preemption, evicting lower‑priority pods to schedule the new one.

Pod Priority

Priority Classes let you assign numeric priority values to pods, making higher‑priority pods less likely to be evicted.

$ kubectl get priorityclasses
$ kubectl get pc

NAME                     VALUE        GLOBAL-DEFAULT   AGE
system-cluster-critical 2000000000   false            2d
system-node-critical    2000001000   false            2d

Priority Example

Three pods represent blueberry, raspberry, and strawberry. Two priority classes are defined:

apiVersion: scheduling.k8s.io/v1
kind: PriorityClass
metadata:
  name: trueberry
value: 1000000
globalDefault: false
description: "This fruit is a true berry"

apiVersion: scheduling.k8s.io/v1
kind: PriorityClass
metadata:
  name: falseberry
value: 5000
globalDefault: false
description: "This fruit is a false berry"

Blueberry gets trueberry (value = 1000000)

Raspberry and strawberry get falseberry (value = 5000)

When new high‑priority pods are created, lower‑priority pods (raspberry, strawberry) are evicted to free resources.

priorityClassName: trueberry

Node‑Pressure Eviction

Kubernetes monitors node resources such as CPU, memory, and disk. When usage exceeds defined thresholds, the kubelet evicts pods based on their QoS class.

QoS Classes

Guaranteed

Burstable

BestEffort

QoS is determined by the pod’s CPU and memory limits and requests.

Guaranteed

All containers set both limits and requests.

CPU limits equal CPU requests.

Memory limits equal memory requests.

Guaranteed pods are rarely evicted.

Burstable

Not Guaranteed.

At least one container has a limit or request.

Burstable pods can be evicted, but less likely than BestEffort.

BestEffort

No container sets limits or requests.

BestEffort pods are the most likely to be evicted under node pressure.

Kubelet evicts pods in this order:

BestEffort or Burstable pods that exceed their requests.

Burstable pods below requests or Guaranteed pods.

Other Eviction Types

API‑Driven Eviction

You can request eviction via the Kubernetes Eviction API.

Taint‑Based Eviction

Applying a NoExecute taint to a node evicts pods that do not tolerate it.

Node‑Level Drain

Commands kubectl cordon and kubectl drain prevent new pods and evict existing pods from a node.

Monitoring Pod Evictions with Prometheus

Use the following query to see evicted pods:

kube_pod_status_reason{reason="Evicted"} > 0

Combine with kube_pod_status_phase{phase="Failed"} to identify pods that failed before eviction.

Conclusion

During preemption, Kubernetes evicts lower‑priority pods to schedule new ones, using Priority Classes to control eviction likelihood. Node‑pressure eviction respects QoS classes, allowing you to protect critical workloads by configuring appropriate limits, requests, and priorities.