Why we need odd masters and even workers nodes, is there any rationale behind it . . .

Kubernetes
#1

ameya agashe:
Why we need odd masters and even workers nodes, is there any rationale behind it ?

#2

OE:
Yes. It’s a distributed solution and most control plane nodes need to have a leader (etcd and basically everything that has an active-passive configuration such as the various controllers). You need an odd # to resolve elections as you could run into deadlocks if you have an even number

#3

runlevl4:
You’ll find this in a variety of scenarios in this space. For example, we’re in the process of deploying CockroachDB. For a true high-availability (HA) solution it needs a quorum to address the same type of leader election. We only have two datacenters so we’re having to deploy to AWS as well.

I haven’t seen any requirements on odd or even workers. We don’t follow that. Our clusters are currently even but that’s only because we added a node to each. They’ve been odd numbers since day one.

#4

ameya agashe:
@OE Thank you mate.
@runlevl4 So you saying your K8S Cluster was Odd masters and now its even masters ?

#5

runlevl4:
No. Three masters and x workers. I was pointing out that we don’t run even workers as a rule. We run the number we need to meet the load.

#6

ameya agashe:
Ok Sir, got it