A naukri.com initiative
The New Stack
7d
103
Image Credit: The New Stack
How K8s CPU Requests and Limits Actually Work — Chapter 2
- To pick up again, imagine that in your Kubernetes pod spec, you asked for 250 millicores of CPU to run your container.
- Something has to happen to turn that abstract request, 250m of CPU, along with any limits, into a set of concrete allocations or constraints around a running process.
- Most Kubernetes resource abstractions are implemented by kubelet, and the container runtime, using Linux Control Groups (cgroups) and control group settings.
- When Kubernetes sets cpu.max (limits), that doesn’t change the process’s proportional priority while it’s runnable.
- There is often moment-to-moment spare CPU capacity on a node that isn’t guaranteed to a particular container by virtue of its CPU requests.
- The limits approach can feel tempting at first, but an industry consensus has been building around not using CPU limits at all for general-purpose workload templates and instead relying on the request method.
- People usually intuit that limits are related to fairness and make sure every workload gets its allotted time.
- With CPU and CFS cgroup settings out of the way, it’s time to move on to memory.
- How do memory requests and limits translate to Linux process settings?
- Will proportionality modeling apply to memory the same way it does to CPU?
Read Full Article
6 Likes
For uninterrupted reading, download the app