Skip to content

Core Concepts

A short tour of the nouns you'll meet throughout DC Suite.

Cluster

A cluster is a set of machines DC Suite reserves, provisions, and wires together for you. It has a lifecycle:

stateDiagram-v2
    [*] --> PENDING
    PENDING --> ALLOCATING
    ALLOCATING --> PROVISIONING
    PROVISIONING --> CONFIGURING
    CONFIGURING --> READY
    READY --> READY_DEGRADED
    READY_DEGRADED --> READY
    READY --> DELETING
    READY_DEGRADED --> DELETING
    DELETING --> DELETED
    ALLOCATING --> FAILED
    PROVISIONING --> FAILED
    CONFIGURING --> FAILED
  • READY — fully formed and usable.
  • READY_DEGRADED — usable but one or more nodes need attention.
  • FAILED — provisioning could not complete (the detail page explains why).
  • DELETED — torn down; the record is kept for history and cost.

Node

A node is one machine in a cluster. Nodes come from the fleet's pool and carry a hardware profile. Within the fleet a node has its own lifecycle — AVAILABLE, RESERVED, IN_CLUSTER, SANITIZING, QUARANTINE — but as a user you mostly care that a cluster's nodes are up and reachable.

Hardware profile

A profile describes a class of machine — GPU model, GPUs per node, allocation unit, interconnect. Examples: gpu-l40s, gpu-hgx-h100. When you create a cluster you choose a profile and a node count; the profile picker shows how many nodes are available right now so you only ask for what the fleet can satisfy.

A single cluster can mix profiles as separate node groups — for example a few H100 nodes for training plus CPU nodes for pre-processing. This is a heterogeneous cluster.

Image

An image is the golden machine image nodes boot from. Images are versioned and move through a lifecycle:

BUILT → CANARY → STABLE → DEPRECATED

Clusters default to the current STABLE image; you can pin a specific version at creation time. See Images.

Template (stack)

A template (also called a stack) is a packaged piece of software you apply to a running cluster — a scheduler, an ML framework, a dashboard. Templates can be global (published by an admin, visible to everyone) or private (yours only). See Templates.

Tenant and organization

  • An organization is the top-level account boundary. Cost is rolled up per organization, and access is scoped by it.
  • A tenant is a project/team within an organization, with its own quotas (max nodes, max clusters) and its own clusters.

Your identity belongs to an organization and typically a tenant; you only see and act on resources within your scope.

Roles and permissions

DC Suite uses scoped roles. A role grants a set of permissions (cluster:write, billing:write, baremetal:read, …) at a scope (installation, organization, or tenant). If an action is refused or a control is hidden, you lack the permission at that scope — ask an administrator.

Operations

Long-running actions (create, resize, delete) run as asynchronous operations with their own status. The console tracks them for you; via the API you poll the operation until it completes. See API Reference.

Next: put it to work in Creating a Cluster.