What is a cluster?

What it means

Each server in a cluster is called a node. The nodes talk to each other constantly and share the work. When a node stops responding, the rest notice and restart its workloads on another node. This is called high availability. Because the workloads are restarted, users see a short interruption while it happens.

To avoid two halves of a broken cluster each believing they are in charge, the nodes use quorum. Quorum means more than half of the nodes must agree before the cluster acts, which is why three nodes is the usual minimum.

Workloads can only move if every node can reach their data. That needs shared storage, or storage replicated between nodes. The idea applies to Proxmox VE, to Ceph storage and to Kubernetes. It is what lets you move a virtual machine between servers, as described under virtualization.

Why it matters to a business

• Survives a failure: a dead server or a power supply fault costs a few minutes while workloads restart, instead of an outage that lasts until the hardware is repaired.
• Planned maintenance: you can update one node at a time while the others keep working.
• Room to grow: adding a node adds capacity without replacing the whole system.
• What it costs: a cluster needs more hardware, a good network and more careful design. It also does not replace backups, because a mistake or an attack is copied across every node.

How this shows up in our work

We build Proxmox VE clusters with high availability and live migration, and Ceph clusters that provide shared storage for them. We also build Kubernetes on Talos Linux. Our own infrastructure runs on Proxmox and Ceph. Virtual machines on the clusters we build are backed up with Proxmox Backup Server, and restores are tested.

Learn more

• Computer cluster on WikipediaA neutral overview of how clusters are built and used, helpful for general readers.
• High-availability cluster on WikipediaExplains failover and why redundancy keeps services running, for readers weighing uptime needs.
• Proxmox VE documentationThe official chapters on clusters and high availability, written for administrators.