Glossary
Short definitions of the terms used across the Clustron documentation. Each entry matches how the concept pages define the term.
Attach mode — A deployment model in which an external orchestrator (Docker, Docker Compose, or Kubernetes) runs the node processes and the manager only observes them while serving the control plane and console. Selected with ZARIS_MGMT_MODE=attach; in this mode the manager refuses to fork stores.
Cluster — The set of nodes that together form and run one store. The nodes exchange membership over their cluster port, elect a leader, and present themselves to clients as a single store.
Control plane — The part of Zaris that manages the cluster itself rather than the data in it: store creation, node lifecycle, configuration, and port management. It runs when stores and instances are created or torn down, not on every request.
Data plane — The request path that touches your data — key-value operations, prefix queries, watch subscriptions, TTL enforcement, leases, locks, and counters. It runs on every operation and is tuned for throughput.
Key-value store — A store that keeps data as keys mapped to values, which you read and write by key. In Zaris every higher-level feature — TTL, leases, and coordination — is built on top of key-value pairs.
Lease — Ownership with a limited lifetime. One client holds a resource for as long as it keeps renewing the lease, and if that client fails the lease expires and the resource is released. Leases are the foundation for coordination in Zaris.
Lock — A coordination primitive that gives one client exclusive access to a resource at a time. Zaris locks are lease-backed, so a crashed owner's lock is released automatically and cannot deadlock everyone else.
Manager — The Zaris management service that a server runs on port 7801. It forms the control plane, stays off the request path, and is a machine's point of administration. Also called the management service.
Node — A single running store process. A cluster is made of nodes; each holds one partition-replica slot and listens on a client port and a cluster port.
Nodus — The Clustron clustering and messaging core. It provides peer-to-peer membership, leader election, typed messaging, and failure detection with no external coordinator. Zaris clusters on it.
Partition — A slice of a store's keyspace. Keys hash into partitions, each partition is owned by one node at a time, and each carries a stable code-word name (Amber, Onyx, …) so you can follow it across the cluster.
Primary — The authoritative copy of a partition, held by one node. Writes always go to the primary, and reads go to it by default. If the primary's node is lost, one of its replicas is promoted to take over.
Replica — A copy of a partition kept on a different node for availability. A replica receives the primary's writes, and copies of the same partition are always placed on different machines, so losing one machine cannot take out both a primary and its replica.
Replication factor — The number of copies of each partition that exist — the primary plus its replicas. It is the main dial for availability under node loss, and you set it explicitly when you create a store (New-ZrStore -ReplicationFactor).
Server — A machine that runs Zaris, identified by its management address (host:7801). It hosts the manager, and in supervisor mode it forks and owns the node processes on that machine. The server is the unit of scaling.
Store — The central logical unit you work against. Every data and coordination operation runs inside a store, which defines the boundary for consistency, isolation, and lifecycle. The same store abstraction runs in-process or across many nodes.
Supervisor mode — A deployment model in which the manager creates and owns the nodes, forking, starting, stopping, and scaling them. It is the default (ZARIS_MGMT_MODE=supervisor) and the model for hosts you manage directly, such as Windows services and VMs.
TTL (time-to-live) — A duration attached to a key that sets how long it should exist. When the TTL elapses the key is removed automatically, giving you time-based cleanup without any cleanup code. TTL controls the lifetime of data, not ownership of a resource.
Watch — A coordination primitive that lets a client subscribe to changes on keys or key prefixes and receive push notifications the moment they happen, so an application reacts instead of polling.
Workspace — A client-side named connection to a set of managers. It records the manager endpoints — plus an optional admin token and the cluster's deployment mode — that the CLI and web console target. It is not a server, a store, or a cluster.
Zaris — The flagship Clustron product: a distributed, in-memory key-value store with built-in coordination primitives (locks, leases, counters, watches, and transactions). It clusters on Nodus, with built-in security (mutual TLS, offline-verified bearer tokens, and RBAC) and built-in metrics.