I'm a bit confused and could use some clarity on this. If I have just one Kubernetes cluster running in a single location, can I still say that it's high availability (HA)? Or do I really need a second cluster in another location—like a disaster recovery setup—to claim that I have HA? What does everyone think?
5 Answers
Honestly, it depends on a lot of factors. To really boost uptime, you need to automate your monitoring and tracking first. Having clusters in two different regions is a solid step towards HA, and depending on your needs, you might even want to go multi-cloud. But remember, it's less about the infrastructure and more about the reliability of your workloads.
Just remember, one is none and two is one! If you have at least two of everything in one location, I’d call that high availability. How far you take it in terms of adding more sites really depends on your desired level of redundancy.
It really boils down to your reliability needs. Having your pods spread across multiple nodes can give you some level of high availability, but you're still exposed to regional or zonal outages since everything is in one site. So if one region goes down, you're out of luck. If you want true HA, consider having at least two clusters in separate locations.
High availability typically means having multiple data centers within the same geographic region, spaced around 70 to 150 km apart. If you're on a bare metal Kubernetes cluster, that's something worth considering. Just keep in mind the configuration and capabilities of your cloud provider.
What are your specific availability requirements? A single cloud region generally gets you about 99.9% uptime, which is considered HA since you've got multiple nodes or zones. However, you'd still have risks like single points of failure if, for instance, your control plane goes down due to an etcd crash. That's something to consider.
Yeah, I totally agree! It really does depend on what level of reliability you're aiming for. There are many approaches to enhance reliability by using multiple replicas, nodes, or even clusters across different regions.