Kubernetes Pod Eviction: Causes and Prevention

💡 Quick Answer: Understand why Kubernetes evicts pods and how to prevent it. Covers resource pressure, priority classes, PDBs, and eviction policies.

The Problem

This is one of the most searched Kubernetes topics. A comprehensive, well-structured guide helps engineers of all levels quickly find actionable solutions.

The Solution

Detailed implementation with production-ready examples below.

Why Pods Get Evicted

# Check eviction events
kubectl get events --field-selector reason=Evicted -A
kubectl describe pod <evicted-pod> | grep -A5 "Status\|Reason\|Message"

QoS and Eviction Order

# Guaranteed (last to evict) — requests == limits
resources:
  requests:
    cpu: 500m
    memory: 256Mi
  limits:
    cpu: 500m
    memory: 256Mi

# BestEffort (first to evict) — no requests/limits
# DON'T do this for production workloads

Priority Classes

apiVersion: scheduling.k8s.io/v1
kind: PriorityClass
metadata:
  name: critical-apps
value: 1000000
globalDefault: false
preemptionPolicy: PreemptLowerPriority
description: "Critical production workloads"
---
# Use in pod spec
spec:
  priorityClassName: critical-apps

graph TD
    A[Node under memory pressure] --> B{Eviction order}
    B -->|1st| C[BestEffort pods]
    B -->|2nd| D[Burstable pods exceeding requests]
    B -->|3rd| E[Guaranteed pods - last resort]

Frequently Asked Questions

How do I prevent eviction?

Set Guaranteed QoS (requests == limits), use PodDisruptionBudgets, and use PriorityClasses. For node pressure eviction, ensure nodes have enough capacity.

Common Issues

Check kubectl describe and kubectl get events first — most issues have clear error messages pointing to the root cause.

Best Practices

• Follow least privilege — only grant the access that’s needed
• Test in staging before applying to production
• Monitor and alert on key metrics
• Document your runbooks for the team

Key Takeaways

• Essential knowledge for Kubernetes operations
• Start simple and evolve your approach
• Automation reduces human error
• Share knowledge with your team