Multi-cluster Kubernetes management is software that sets up, updates and watches many Kubernetes clusters from one place, whether they run in a public cloud, in a company's own data centre or at small edge sites. The one place is called a control plane. It is also called Kubernetes fleet management.
A cluster is a group of machines on which Kubernetes runs containers. Small companies have one or two clusters. Large ones have many: one for each team, each country and each cloud provider, and often one in every shop, factory or telecom site. Each cluster needs the same care. Someone has to install it, keep its version up to date, apply security settings and install the same tools on it. With 400 clusters, doing this by hand, one cluster at a time, is impossible.
The software works from a template. An engineer describes, once, what every cluster of a given kind should contain: the Kubernetes version, the operating system, the network settings and the apps. The software creates clusters from this template, and then keeps checking each one. If a cluster drifts away from the template, for example because someone changed a setting by hand, the software reports it or puts it back. To update the clusters, the engineer changes the template, and the software rolls the change out in waves. If the first wave has problems, the rollout stops before it reaches the rest.
Consider a supermarket chain with 450 stores across South India. Each store has a cluster of three small computers in a back room. The cluster runs the billing counters, the stock app and the screens that show shelf prices. Because these apps run inside the store, billing continues even if the store's internet link fails. Running apps close to where they are used, like this, is called edge computing. The chain also has two clusters in a public cloud for its online orders. Now the IT team must apply a security update to all 452 clusters. It updates the template, and rolls it out at night. First it goes to 10 stores. After a day without problems, it goes to 100 more. Then it goes to the remaining 340 stores and the two cloud clusters. Three stores fail to update, because their internet link is down. The software shows them in red, and retries them the next night. The whole rollout takes four nights, and needs four engineers.
A company that mixes its own data centres with public clouds has a hybrid cloud, and fleet management lets it treat all its clusters in the same way. Cloud operators use it too. Private cloud software often creates clusters for its customers and hands them to fleet management to run.
The Kubernetes platform page covers what happens inside one cluster. Fleet management is about keeping hundreds of clusters the same.