How Many Servers to Use for a Cluster
Servers used in a cluster and performing the "Cluster component" role (see Roles) are called nodes. This term is used both in the web interface and in the documentation when discussing clusters.
Choosing Between Cost and Resilience
In accordance with the SHIPOPS principle, in addition to scaling as a basic cluster function, two competing goals are considered: saving resources and providing fault tolerance. The user must decide which is more important.
A cost-saving but not fault-tolerant cluster consists of one leader node, with the ability to add new nodes as load grows.
A fault-tolerant cluster is created with several nodes from the start. The number of nodes must be odd and at least three. The ability to add new nodes is preserved.
If the project is enterprise-level, it is better to implement it using several specialized clusters instead of one. In this case, not all of them will require fault tolerance. This approach saves resources even in large projects by combining fault-tolerant and cost-saving clusters.
Common Starting Points
If the project is a test project, a startup with a small audience, or part of a large project without fault-tolerance requirements, a cost-saving one-node cluster is usually enough. If fault tolerance is important, choose a cluster with an odd number of nodes, at least three.