Workers

Immutable app identifier, unique among the authenticated organization's live apps.

Maximum number of items to return.

Opaque pagination cursor returned by a previous call, as nextCursor or, on the operations that offer one, prevCursor.

Scope the page to one version. The default is the app's activeVersionId. When that field is unset the default page is empty: the app has no active version, not that it has no workers. Send all (any case) to include every version. An empty value is refused. A cursor must be replayed under the same version scope it was issued with.

Narrow the page by worker state. The default, all, keeps the terminal stopped rows in the page. live drops them. A state of live with a status of stopped is a contradiction and is refused, because an empty page would read as "this app has never run".

Worker lifecycle status. Also the type of UsageEvent.eventType, which records a ledger subset of these states (see that field: busy never appears there). unhealthy means the pod exists but failed to become or stay ready, not an intentional drain or stop.

Case-insensitive literal substring match against id, podName, nodeName and versionId. A worker matching any field is returned within the selected version, state and status filters. Omit q to disable search. Cursors must retain the same search term, ignoring case.

Cursor for the next page. Null when there are no more items.

Immutable app identifier, unique among the authenticated organization's live apps.

Worker identifier, the Kubernetes pod UID recorded by the reconciler. This is the stable key to address a worker with. podName is the value to show a person.

Immutable app identifier. Unique among the authenticated organization's live apps: it cannot be changed after creation, and it becomes available again once the app it named reaches deleted.

Kubernetes node the pod is scheduled on. Omitted while still pending.

Worker lifecycle status. Also the type of UsageEvent.eventType, which records a ledger subset of these states (see that field: busy never appears there). unhealthy means the pod exists but failed to become or stay ready, not an intentional drain or stop.

Kubernetes-facing reason for the current status when unhealthy or otherwise notable (e.g. ImagePullBackOff, CrashLoopBackOff). A pending worker the cluster could not place reports Unschedulable, which means no node could give it the GPUs the app asks for.

When the worker entered its current status. For a stopped worker this is when it stopped, so it is also the end of the worker's uptime.

GPUs attached to this worker at observation time. Zero means the snapshot is unknown: the worker is still unscheduled, or a terminal observation could not read a GPU pair. A known snapshot is >= 1 and travels with gpuType. An empty pair is incomplete, not a CPU worker. CPU workloads are not supported.

GPU catalog code snapshotted at observation time. Omitted until a GPU snapshot exists (the worker is still unscheduled, or the observation could not read a type). Not a CPU-worker marker. CPU workloads are not supported.

How the catalog currently provisions this worker's gpuType. Omitted when there is no gpuType, or when the catalog no longer holds the code. Unlike gpuType this is read now rather than snapshotted, so it tells you how that GPU is supplied today, not how it was supplied when the worker started.

Worker-reported liveness. Omitted until the worker reports. Stays absent until a heartbeat path exists (phase 2).

Pod creation time (metadata.creationTimestamp), not insert time. It is also the start of the worker's uptime.