Kubelite takes 1.5GB for a 16GB machine

Summary

I have a 16GB machine which installed microk8s. However, kubelite itself takes 1.5GB, which seems a lot considering the total memory is so limited.

Is it a bug with microk8s, or is microk8s just this memory-hungry? Thanks.

What Should Happen Instead?

less memory

Reproduction Steps

1. n/s

Introspection Report

[cannot post here for privacy, can you give an email so I will send it]

Can you suggest a fix?

willing to pr

Are you interested in contributing with a fix?

yes

[ktsakalozos]

Hi @fzyzcjy,

What version of Kubernetes are you using? What workloads are you running?

Kubelite runs all the Kubernetes services started from a single process. This includes the control plane unless you microk8s join the nodes with the --worker flag. In past MicroK8s versions Kubelite also packs the datastore used by K8s.

Here are some configuration options to reduce k8s memory footprint:

• If you are not interested in high availability call microk8s disable ha-cluster immediately after you snap installed MicroK8s.
• If you have a multinode cluster join the nodes as worker nodes via the --worker flag
• if you know you would not be using some parts of the API offered by the API server, use the --runtime-config argument [1] to not load the APIs you do not use. THis argument should be added in /var/snap/microk8s/current/args/kube-apiserver.

[1] https://kubernetes.io/docs/reference/command-line-tools-reference/kube-apiserver/

[fzyzcjy]

Thank you!

(Btw I was on 1.21 because 1.22 has a lot of API changes…)