Hello, asking for a bit of crowd wisdom - what’s t...
# rke2
b
Hello, asking for a bit of crowd wisdom - what’s the fastest way for provisioning a rke2 cluster in HA deployment? I.e. shortest time from 0 nodes to e.g. 10. I want to keep control plane/etcd separate from data nodes. K8s cluster is created with VMs, so there is some time on VM creation/initialisation too. So far I have: • 1. step - 1 ctrl/etcd, 1 worker - to get cluster up and running via rancher custom provisioner • 2. step - 1 more ctrl/etcd - triggers reconciliation • 3. step - 1 more ctrl/etcd - triggers reconciliation • 4. step - all remaining workers Is there a better way?
n
If your Rancher hosted cluster has decent hardware (ie isn't something like a Rasberry Pi) you should be just fine to simplify the setup to 1. 3 ctrl/etcd, 1 agent 2. all remaining agents (workers)
For scale testing, we generally recommend adding agents in batches 50, purely to prevent the Control plane (CP) nodes from needing more memory than they would at a steady state.
If you don't care about the memory spikes and have plenty of excess RAM on the CP, you can add agents in batches of 100+
🙌 1
b
Thanks! Adding 3 ctrl at once - this has been a pretty fragile operation with rke1, often leading to a stuck cluster. I haven't yet seen the same with rke2, but just for safety thought of a sequential addition of etcds. Too safe?