https://rancher.com/ logo
Join Slack
Powered by
This message was deleted.
# k3s
a
This message was deleted.
c
why did you do that? you don’t need 7 control-plane nodes. I would probably do 3 control-plane+etcd and 4 worker (agent) nodes.
m
Okay. That was just an example (sorry for the bad example). What I would like to know is if a node with etcd goes down can that role be assigned to some other node within the cluster to maintain the total number of etcd instances?
c
no. you select the roles when you create the server. Roles are not dynamically assigned.
If you want HA for etcd you should have 3 or 5 etcd nodes. with 3 you can tolerate a single node outage, with 5 you can tolerate 2.
For control-plane there is no quorum requirement, so you don’t generally need more than 2 for HA.
So folks usually do 3 etcd + 2 control-plane + N worker(agent) nodes, or just 3 etcd+control-plane + N worker(agent)
m
Got it. Thanks for the details. And yes you are right. generally 2 control-plane nodes are fine.
The reason I am asking this question is- this is handled differently in Kubernetes, these components are decoupled and therefore can be managed automatically.
I was looking for something that maintains the minimum number of etcd instances running in the cluster.
c
what Kubernetes distro are you looking at, that automatically adds etcd nodes to the cluster as necessary?
m
Oh, no I meant since control-plane and etcd are decoupled, one of them going down wouldn't impact the other.
But i just realized even these components can be decoupled in k3s
2 Views
Open in Slack
PreviousNext
Powered by