Infrastructure & Operations › Cloud Computing

Cluster

A group of machines working together as one system.

Also known as: compute cluster, kubernetes cluster, cluster of servers

A cluster is a group of machines treated as one system. One machine has a ceiling — cores, memory, and a single point of failure — so when a workload outgrows it, you join several together and let software distribute the work. The word covers two related ideas: a general compute cluster, and a Kubernetes cluster specifically.

In generic terms, a cluster pools servers or compute instances behind a scheduler that decides what runs where. That gives you more capacity than any one machine and, if designed well, survives the loss of a node.

A Kubernetes cluster is the common case today: a control plane manages a set of worker nodes that run pods. Managed offerings run the control plane for you, so you mostly deal with nodes and workloads. See container orchestration.

The classic mistake is a “cluster” of one node. It has the same single point of failure as a single server, but with more moving parts to operate. If you’re building a cluster for availability, spread the nodes across failure domains — different machines, ideally different availability zones — and make sure the workload can actually run on more than one node.

A second mistake is assuming a cluster is automatically fault-tolerant. It isn’t, unless things are redundant: a database on one node is still a single point of failure no matter how many app nodes exist.

When not to use it: if one server handles the load with room to spare, that’s simpler and cheaper. Clusters pay off for scale, resilience, or when the orchestrator’s features (rolling updates, self-healing, autoscaling) earn their keep. Very large organizations sometimes run several clusters divided by team or environment — see multi-cluster.