Managed K8s (control plane + workers)
Some customers don't want to nurse individual servers -- they want a cluster that automatically restarts their apps when a machine breaks. Managed Kubernetes gives them exactly that: a group of worker machines plus a "control plane" brain that keeps their apps running across them. You bill a flat fee for the control plane and an hourly rate for each worker.
The cluster's health is a fraction of its workers. Lose one worker and Kubernetes shuffles the apps onto the survivors -- the customer barely notices. Lose a third of them and the cluster is officially degraded; lose more and it's down. So spreading those workers across different physical servers really matters -- the blast radius of one dead server is the whole point of buying redundancy.
Detailed explanation
K8sDemand provisions a control plane (flat ~$73/mo) plus N worker hosts pinned one-per-host for anti-affinity. Per-tick billing prorates the monthly control-plane fee + a per-worker hourly rate. An ingress load balancer typically fronts the cluster, with backend count tracking node count.
Cluster is healthy at >=2/3 nodes up, degraded between 1/3 and 2/3, down below 1/3 -- host failure proportionally craters cluster capacity. Because workers are anti-affinity-spread one-per-host, a single host loss drops exactly one node; the danger is correlated loss (a rack breaker trip or shared-switch outage taking several at once), which is why these tenants care about the host blast radius.