Skip to content

Capacity and placement

Pool. A pool is the placement boundary selected by a production or runtime. It groups capacity with a common operational purpose, such as an on-premises cluster or a cloud region.

Tier. A pool contains an ordered list of tiers, from the capacity you would prefer to use to more expensive fallbacks. A tier says where capacity comes from and which machine types are candidates. Tiers can use AWS, OCI or registered local servers. The ordered tier list is one value: changing it replaces the whole list.

Cloud. A cloud is a named provider configuration containing the region, network, worker image and other facts needed to create nodes. Credentials remain outside this resource. A cloud-backed tier refers to it when the Manager needs to provision capacity.

Machine type. A machine type describes a capacity shape that a cloud tier may provision, including the resources and capabilities it offers. The catalog can contain shipped and operator-defined shapes.

Node. A node is one unit of worker capacity known to the Manager. It reports its inventory and health, hosts runtime executions, and has lifecycle controls such as cordon and drain. A cloud node is provisioned from a cloud tier; a local node is backed by a local server registration.

Local server. A local server is a pre-existing machine registered into a local-server tier with a registration token. The token is shown once, when the server is registered, and the Manager keeps only its hash; the server presents it each time it (re)registers. While a local server is connected, a registration from a different host is refused, so a copy of the token cannot move it elsewhere. norsk-mgr local-servers list shows every registration until it is deleted; --reachable narrows it to the servers reporting in now. Its registration and resulting node are the same operational identity. Deleting it removes that identity and token; using the machine again means reprovisioning it.

Requirements, capabilities and inventory. A runtime’s requirements describe what it needs. A node’s capabilities and current inventory describe what it can provide. The Manager compares them, along with pool, tier and availability, before assigning work.

Placement. A placement is the diagnostic record of that decision: which nodes were considered, why candidates were accepted or rejected, and where the runtime ended up. It is history and explanation rather than something an operator creates or edits.