kube-apiserver: very high memory load in Raspberry PI

Hi,

I’m running a freshly installed MicroK8s cluster v1.20 on top of Ubuntu Server 20.04 on 3 Raspberry PI v4. After a couple of hours I noticed that some of my pods are not responing anymore and kubectl doesn’t return anything either. A look onto the system reveled the issue: full memory and the process kube-apiserver took around 70-80 % of it.

At the moment I execute sudo snap restart microk8s.daemon-apiserver every hour (cronjob) to keep the cluster somehow alive.

Recently I switched from K3s to MicroK8s due to high cpu and memory load in K3s. Maybe there is an overall issue with K8s on top of Raspberry PIs?

Targerball of microk8s inspect: inspection-report-20210115_091600.tar.gz

[Ramblurr]

For folks landing here from Google, this seems to be the current summary:

• There’s a memory leak when HA mode is enabled
• There is no solution or workaround that lets you keep using HA mode
• Disable HA mode might fix it, but you might have to purge and reinstall

[balchua]

To those who are interested, tthere are few fixes on dqlite. The fixes will be backported to 1.19. and 1.20.

My personal test so far is promising.

https://github.com/ubuntu/microk8s/issues/1598#issuecomment-779044844