Kubernetes Resource Limits and Requests Best Practices

Pods without proper resource requests/limits get evicted unexpectedly, starve other pods, or waste cluster resources.

Always set both requests and limits:

resources:
  requests:
    memory: '256Mi'
    cpu: '250m'
  limits:
    memory: '512Mi'
    cpu: '500m'

Guidelines:

1. Requests = guaranteed minimum. Scheduler uses this for placement
2. Limits = hard ceiling. Container gets OOMKilled (memory) or throttled (CPU)
3. Set requests based on actual usage (check metrics)
4. Set memory limit = 1.5-2x request (OOM is fatal)
5. CPU limits are debatable - throttling causes latency spikes. Consider no CPU limit with just requests
6. Use LimitRange for namespace defaults
7. Use ResourceQuota to prevent one team from hogging cluster

Quality of Service classes:

• Guaranteed: requests == limits (highest priority)
• Burstable: requests < limits
• BestEffort: no requests/limits (first to be evicted)

Without requests, the scheduler cannot make informed placement decisions. Without limits, a single misbehaving pod can take down a node.

Deploying applications to Kubernetes clusters