Network partition between 2 nodes in HA setup causes reachable node's API not to respond

[renannprado]

I have executed this setup, so basically I have this:

NAME       STATUS   ROLES                       AGE     VERSION
node1   Ready    control-plane,etcd,master   2d13h   v1.24.4+k3s1
node2   Ready    control-plane,etcd,master   15h     v1.24.4+k3s1

node1 is in my home network, while node2 is in the cloud, they are masters as well as workers 😄 . Both nodes are connected through wireguard.
I wanted to test what happens if there's a network partition, so I switched off the wireguard connection between the two.

After doing that, when I run kubectl get pods I get:

Get "https://192.168.169.130:6443/api/v1/namespaces/default/pods?limit=500": dial tcp 192.168.169.130:6443: connect: connection refused - error from a previous attempt: unexpected EOF

Even though the port on the node1 is still reachable:

# nc -vz 192.168.169.130 6443
Connection to 192.168.169.130 port 6443 [tcp/sun-sr-https] succeeded!

I know I should be setting up 3 nodes minimum, but I would like to stick with 1 powerful node at home instead of two smaller ones, that's a conscious choice. If that's a cause of this issue, then I would consider having a 3 node setup, otherwise I would like to resolve it so that this doesn't happen when, for example, I don't have internet at home, therefore killing the connection between the two nodes.

[brandond]

Duplicate of #5210 - see #5210 (comment)

tl;dr you need an odd number of etcd nodes for quorum. Two nodes gets you pretty much nothing.