| Deployment | Deployment | A Deployment keeps a replicated pod template running and can roll changes across replicas. |
| Pod | Pod | A Pod runs one or more containers as a single scheduled unit. |
| ReplicaSet | ReplicaSet | A ReplicaSet keeps a fixed number of pods running from one pod template. |
| DaemonSet | DaemonSet | A DaemonSet keeps one pod running on every eligible node of its subnet.
A node is eligible when it is admitted to the DaemonSet’s subnet, belongs to the fleet the pod template’s placement names (when it names one), is capacity the DaemonSet’s account may run on, and advertises what the pod template needs. The pod count follows that set as nodes join and leave; the manifest states no replica count. Each pod is bound to its node for life: a failed pod is replaced on the same node and is never placed on another. |
| HostPodSet | HostPodSet | A HostPodSet keeps a counted set of host Pods. A host Pod runs no workload of its own: it reserves a CPU and memory envelope on a node for isolated units that a node agent creates inside it later, through the node’s own contract. The HostPodSet decides how many hosts exist, where they may be placed, and the isolation class of the units they hold; it says nothing about what those units run or for whom beyond its opaque owner.
A HostPodSet is reconciled like a ReplicaSet: the platform creates and removes host Pods until the placed set matches the desired count, and when the template changes it replaces hosts according to the strategy, draining each host before it is removed. |
| Job | Job | A Job runs a pod template until the requested completions succeed. |
| CronJob | CronJob | A CronJob creates Jobs on a schedule. |
| Sandbox | Sandbox | A Sandbox runs a temporary pod template until its TTL expires.
A Sandbox stays available until its TTL expires or it is deleted. It runs one Pod at a time. When the Node running that Pod drains, for an update or because the Node is being deleted, IDYL stops the Pod gracefully and starts a fresh Pod from the template on a Node the Sandbox may run on, which can be the same Node once its drain has ended; the Sandbox reports pending until the fresh Pod runs. Nothing written inside the stopped Pod carries over. A Pod that ends any other way ends the Sandbox, which reports failed. |
