HA Cluster Lab exercise to be verified

StefanDiaconu

Hello,

on the September 2024 version of the LFS258 course, the presented implementation HA cluster exercise LAB includes several control planes and a single loadbalancer to distribute the load and to have a unique endpoint for all.

After analysing the setup I understand that this is not a High Availability cluster as it gives only 1 loadbalancer and 1 ip address(real ip address of the hostname on which is installed the haproxy) that are exposed to failures, so the HighAvailability principle will be broken as soon as the loadbalancer or the IP address of the endpoint fails.

Please review the lab exercise for HA cluster and provide an example of a real HA implementation that will be persistent even if other components than etcd database will fail - in order to respect the High Availability definition.

chrispokorni

Hi @StefanDiaconu,

You are absolutely correct. Carefully analyzing the setup, one can find that the solution presents a weakness. The lab exercise mirrors the official Kubernetes docs recommendations for setting up a highly available Kubernetes control plane cluster. You can find out more on the topic at:

https://kubernetes.io/docs/setup/production-environment/tools/kubeadm/ha-topology/

Agreed, both solutions presented by the official docs are less than perfect. However, they are a starting point for a truly HA Kubernetes control plane cluster. Instead of a single LB one can deploy a distributed set of redundant LBs, paired with floating IP or virtual IP address, Round-Robin DNS, etc. Clearly there are plenty of implementation options to achieve full HA, all dependent on the infrastructure hosting the solution and the services that implement it.

Regards,
-Chris