Small k8s (not kubernetes-core)


#1

Hello!

For development purposes, I have deployed the canonical-kubernetes bundle on my private MaaS cloud.

I must have made some network setup mistake, since I am not able to deploy any of the core-* bundles while the full 10 machines bundle works as designed. My question is: since I do not need HA, can I just remove units etcd/1, etc/2 and kubernetes-master/1 after deployment?

Of course I’d like to fix the core-* issue when I have more time. I tried some different setup but lxc containers receive the wrong network, or DNS is not working on the host, etc. I have a smaller MaaS cloud with a single network where core-* works.

Thanks


The Juju Show #46 - Machine Learning with Kubeflow March 27th 15:00UTC
#2

I’ll let the @tvansteenburgh and the k8s folks speak directly to how things will perform in a smaller setup but I did want to highlight that microk8s is a great way to do the development side on a local machine or you could set it up on a single node MAAS node if you wanted something that didn’t monkey with your development machine.

https://microk8s.io/

Let me know if that might be useful for you.


#3

Thanks for the hint.
In my use case, I need multiple workers with persistent storage, monitoring stack and possibly a full set of APIs with a configuration as close as possible to production. But since the load is not very high, I though of limiting the resource waste.
I am not at all worried about performance, just wondering about the resilience when I scale etcd and masters down to 1.


#4

For development purposes, yes, scaling everything down to 1 unit will be fine.