Clustering
Clustering is the technique of grouping multiple computers – called nodes – into a single logical system that works together as one. A group of nodes operating this way is called a cluster. To the outside world, a cluster typically appears as a single service, even though the work is shared across several physical or virtual machines.
Clustering is a foundational building block of distributed systems. It is used to raise scalability and fault tolerance, and to keep a service available when individual nodes fail. A cluster gains capacity by adding more nodes, which is a form of horizontal scaling.
Cluster nodes cooperate by coordinating their work. Incoming requests are spread across the nodes using load balancing, often through a dedicated load balancer. Nodes that share state typically rely on replication to keep their data in step, and on consensus algorithms when they must agree on a single value despite concurrent writes or failures.
A common deployment pattern is the active-active cluster, where every node handles live traffic. The alternative is an active-passive arrangement, where a standby node takes over only when the active node fails, a process known as failover. Either way, the goal is that the failure of one node does not bring down the whole service.
Clustering is not free. Coordinating nodes adds communication overhead, and shared state introduces consistency challenges. Adding nodes can expose new bottlenecks, and managing a cluster is more complex than managing a single machine. These costs are the trade-off for the availability and capacity that clustering buys.
Container orchestration systems such as Kubernetes build on clustering. They run containerized workloads across a cluster of machines and automate the deployment, scaling, and recovery of nodes within it.