Loading auth...
Medium10 minContainer Orchestration
UpdatedAug 2, 2026
Edit

Resource Requests vs. Limits

CONCEPTS:Resource Management (Requests vs. Limits)

Why This Is Asked

This is a critical operational question. Poorly defined resources lead to either wasted money (requesting too much) or unstable applications (OOM kills and throttling). Interviewers want to see if you can balance cost and reliability.

Key Concepts

  • Scheduling: How requests affect where Pods land.
  • Throttling vs. Eviction: What happens when limits are hit.
  • OOM Kill: The behavior of the Linux OOM Killer in a containerized environment.
  • Units: Understanding m (millicores) and Mi/Gi (mebibytes/gibibytes).

Answers by Technology

+ Add Variant
KubernetesImprove this answer ✏️

Expected Answer

Requests and Limits are how Kubernetes manages hardware resources for containers.

  • Requests represent what the container is guaranteed to get. The scheduler uses requests to find a node with enough available space. If a Pod requests 2GB of RAM, it will only be placed on a node that has 2GB of unreserved RAM.
  • Limits represent the ceiling. It’s the maximum amount the container can use.
    • If a container hits its CPU limit, it is throttled (slowed down), but it keeps running.
    • If a container hits its Memory limit, it is OOM Killed (Out of Memory) by the kernel and usually restarted by Kubernetes.

Why It Matters

If you don’t set requests, the scheduler might over-provision a node, leading to resource contention. If you don’t set limits, a single buggy container (with a memory leak) can consume all the memory on a node, causing the kubelet to crash or other Pods to be evicted.

Example Code

Resource Specification in a Deployment

spec:
  containers:
  - name: my-app
    image: my-app:latest
    resources:
      requests:
        memory: "256Mi"
        cpu: "500m" # 0.5 cores
      limits:
        memory: "512Mi"
        cpu: "1" # 1 full core

Common Mistakes

  • Setting CPU requests too high: This wastes money because the scheduler reserves that CPU even if the app isn’t using it.
  • Setting Memory limits too low: Causing constant OOM kill loops because the app needs more memory than the limit just to start up (e.g., JVM apps).
  • Not setting any limits: A “BestEffort” Pod is the first one killed when a node runs out of memory.

Follow-up Questions

  • What is a “millicore” (m)? (Answer: 1/1000th of a CPU core. 250m is 25% of a core).
  • What is the difference between MiB and MB? (Answer: MiB (Mebibyte) is base-2 (1024 bytes), while MB (Megabyte) is base-10 (1000 bytes). Kubernetes uses Mi/Gi which are power-of-two units).
  • What happens if a Pod’s memory request is higher than any single node’s total memory? (Answer: The Pod will remain in a Pending state indefinitely because it cannot be scheduled).

References