❌ Myth: If I set cpu: 500m in my pod’s requests, Kubernetes reserves 0.5 CPU exclusively for my pod.
✅ Reality: Kubernetes does not reserve CPU like it does for memory. CPU is a compressible resource, meaning:
-
Your pod gets priority access to 500m CPU but can use more if available.
-
Other pods can borrow unused CPU from your pod if it’s not using the full 500m.
-
If the node is under high load, the kernel’s CPU throttling kicks in, limiting your pod’s CPU usage.
📌 Example:
🔹 You set cpu: 100m request, but your pod only needs 50m most of the time.
🔹 Another pod on the same node borrows the remaining 50m CPU.
🔹 Later, when your pod suddenly needs 100m CPU, it may get throttled if the node is already overloaded.
💡 What to Do?
✅ Set realistic requests—don’t go too high or too low.
✅ Use CPU limits if you want to cap a pod’s max CPU usage.
✅ Monitor throttling using ‘kubectl top pods’ and Prometheus metrics.
