Architecture and operations
This section explains how a Clustron Zaris cluster works as a running system: the processes it is built from, how your data is partitioned and replicated across nodes, the two ways you can run it, how you secure it, and how it grows and shrinks.
Use this section to make capacity, availability, and operational decisions. Where Core concepts answers "what is a store, a key, a lease?", this section answers "what runs on which machine, and what happens on the wire?"
Cluster at a glance
A cluster has two parts. The control plane is a manager process that creates stores, tracks which nodes exist, and backs the web console. The data plane is one or more node processes that hold your data, serve reads and writes, and replicate partitions to each other. Your application is a .NET client that routes each request straight to the node holding the data — the manager is never on the request path.
This split is the theme of the whole section. An administrative operation on the manager cannot stall a hot read on a node, and losing the manager does not lose your data.
In this section
The pages below build on each other. Read cluster architecture first, then choose the deployment model that matches your environment before you tackle security and scaling.
| Page | What it covers |
|---|---|
| Cluster architecture | The manager, the nodes, their ports, and the control-plane / data-plane split. |
| Partitioning and replication | How keys map to partitions, how the replication factor protects against node loss, and what partition code words mean. |
| Deployment models | Supervisor vs attach — who owns the node processes, and how that changes how you operate. |
| Security and TLS | Token authentication and transport TLS, and how to turn each on per deployment model. |
| Elastic scaling | Adding and removing capacity, and why scale-out is lossless while scale-in is not yet drain-safe. |
| Web console | The browser UI for running and observing a cluster. |