What You'll Learn
You'll learn how Kubernetes intentionally allows nodes to promise more CPU and memory than they physically have, and what that means for scheduling and stability.
The Core Idea
Resource overcommitment occurs when the sum of all resource requests across Pods scheduled on a node exceeds the node's actual capacity. Kubernetes allows this by design—the scheduler only considers requests, not limits, when deciding where to place Pods.
Think of it like airlines overbooking flights. They know statistically that not every passenger will show up (not every Pod will use its full request), so they sell more seats than physically exist. This maximizes efficiency but carries risk if everyone actually boards.
How It Works
When you set resource requests for your Pods, the Kubernetes scheduler reserves that capacity conceptually. However:
- Pods with requests lower than limits can burst beyond their request
- Multiple Pods bursting simultaneously can exceed node capacity
- If Pods only specify limits (no requests), the scheduler treats the request as equal to the limit
This means your node might schedule Pods requesting 4GB total memory on a 3GB node if those Pods have higher limits—overcommitting the node.
The Implications
Benefits:
- Better resource utilization (avoiding wasted capacity)
- Cost savings by running more workloads per node
Risks:
- Memory pressure can trigger Pod evictions
- CPU throttling when demand exceeds capacity
- Lower-priority Pods (BestEffort, Burstable QoS) get evicted first
You control overcommitment through LimitRanges, ResourceQuotas, and careful request/limit configuration.
Key Takeaway:
Kubernetes overcommits node resources by scheduling based on requests, not limits—this maximizes efficiency but requires careful capacity planning to avoid resource contention and Pod evictions.