Create a pod

purchase option. on_demand (pay per second, no commitment) is the only class sold; reserved is a deprecated alias that is stored as on_demand. interruptible is retired and rejected with 400 COMPUTE_CLASS_RETIRED; it remains listed only until the deprecation window closes.

how many whole GPUs this pod gets. All of them are claimed on ONE node and passed to the container together; the pod is priced for gpuCount GPUs. Omitted = 1. A count no single node could ever hold for your org is refused with 400 GPU_COUNT_NOT_ALLOCATABLE; one that some node could hold but none has free right now is refused with 503 INSUFFICIENT_GPU_CAPACITY (maxGpuCount on /v1/pods/capacity is the largest a create can take now).

in-guest HTTP port exposed on the endpoint

tenant container image ref (required)

true = open endpoint (no data-plane auth); default false requires an org API key

size of the pod's private /dev/shm in GiB (docker --shm-size). Optional; the maximum is half the pod's memory limit on its placement node (itself the pod's GPU-proportional share of node RAM), and a larger value is refused with 400 POD_ADMISSION_FAILED. Omitted lets the platform pick a safe default: that same maximum for a pod that owns its whole node, 2 GiB per GPU (capped at the maximum) for a shared multi-GPU pod, and docker's default for a shared single-GPU pod. Raise this when a multi-GPU NCCL / PyTorch DataLoader workload needs more shared memory than the default.

size in GB of the pod's persistent volume, mounted read-write at /workspace. Omitted = 50. The volume lives on the pod's node and is kept across stop, start, restart, container crashes, node reboots and interruptions; it is deleted only when the pod is deleted (or after 7 days interrupted with its node gone). Everything outside /workspace is the container's own filesystem and is reset whenever the container is recreated. Fixed at create; there is no resize.

compute-fabric capability tier (e.g. "pod"); empty for a plain VM

compute-fabric purchase option (on_demand; pods created before the rename carry the legacy reserved or interruptible values); empty for a plain VM

GPU vendor from the catalog, the same value as GpuModel.manufacturer (NVIDIA, AMD); absent on a CPU-only row

tenant container image (pods)

platform owner of this row ("serving" = inference serving replica, read-only on customer surfaces); empty for a customer-launched VM

true = open endpoint (no data-plane auth); false = require an org API key

managed-service tag; empty for a plain VM

deploying, running, stopped, failed, terminated, or interrupted; a row can also carry the queued status pending. interrupted means it served and is not serving now: it keeps its node, GPU and disk, and is not billed. A VM is interrupted while its node is offline and resumes when the node returns; a pod is interrupted for that and for losing its supervisor or its job on a live node, which the platform recovers by re-submitting it. After 7 days interrupted a VM is stopped (disk kept) and a pod is deleted.

Cause of the current status. The machine-readable values are "node_lost" (the host went offline and is expected back, so the disk is preserved), "node_removed" (the host was decommissioned, so the disk is gone and a restart deploys fresh elsewhere), "vm_exited", "user_stopped", "user_terminated", "user_restarted", "spot_reclaimed_by_capacity_owner", "spot_preempted_for_reservation_tenant", "auto_stopped_insufficient_balance", "auto_stopped_account_suspended", "auto_stopped_organization_deleted", the prefix "container failed to start: " followed by one sentence of OpenRelay's, and on a running workload the prefix "node leaving at " followed by the RFC 3339 time its host is removed, when the workload ends with node_removed. Anything else is free text for a human to read.

pods only: size in GB of the pod's persistent /workspace volume.