Google Cloud uses the words instance and cluster with different meanings depending on the database service, and that inconsistency is a source of confusion when you move between services. The same term sits at the top of the hierarchy in one service and lower down in another, and in at least one case the two words are used interchangeably. For the Professional Cloud Database Engineer exam, the goal is not to memorize every variation but to recognize which unit a given service treats as its primary deployment resource so the naming in a question does not throw you off. This article walks through how each managed database service uses these terms.
In Cloud SQL, the instance is the primary unit of deployment. Each instance is essentially a virtual machine that runs the database engine. There is no cluster concept layered above it in the way other services define one, so when you provision Cloud SQL you are creating an instance and that instance is the thing you size and manage.
AlloyDB introduces a different hierarchy. The top-level resource is a cluster, which acts as a container for your data. Inside that cluster you create one or more instances, and each instance contains a set of node resources that handle the actual processing. So in AlloyDB the order runs cluster, then instance, then node, with the cluster sitting above the instance rather than the other way around.
With Spanner the terminology shifts again. You start by creating an instance, but here that instance is a container for compute resources rather than a specific server. Those resources are measured in node units, which are distributed to provide Spanner's global scalability. The word instance carries a broader meaning than it does in Cloud SQL, where an instance maps to a single VM.
Bigtable is close to the reverse of AlloyDB. You define a top-level instance, which then contains one or more clusters. Each of those clusters is composed of a specific number of nodes that manage the data processing and storage for a region. So in Bigtable the instance is on top and the cluster sits beneath it, the opposite arrangement from AlloyDB where the cluster is the top-level container.
Memorystore handles naming differently depending on which option you choose. When you use Memorystore for Redis, you are provisioning a single instance. When you use Memorystore for Redis Cluster, the word cluster is part of the service name itself. For that service Google uses the terms cluster and instance interchangeably when referring to a single unit of deployment. A cluster or instance there is composed of several shards, and each shard contains the individual nodes that handle your data.
The takeaway is that instance and cluster do not have a single fixed meaning across Google Cloud databases. In Cloud SQL an instance is a VM. In AlloyDB the cluster is the top-level container and instances live inside it. In Spanner an instance is a pool of compute measured in nodes. In Bigtable the instance is on top and contains clusters. In Memorystore for Redis Cluster the two words refer to the same unit. You do not need to memorize every detail at once, because each service is worth studying on its own. What helps on the Professional Cloud Database Engineer exam is reading the term in the context of the service named in the question rather than assuming it means the same thing everywhere.
Our Professional Cloud Database Engineer course covers this terminology alongside the architecture of each managed database service and how their compute units are sized, with practice questions that drill these distinctions.
