latestOpenAPI 3.1.02026-08-22151184245.5 KB

25f920394c7d

Internal

Pre-deployment checks for a serving spec (platform admin)

Read-only. Evaluates everything that must already be true for a deployment with the given spec to admit, place, and register: the catalog row is seeded with pricing, the weights ref is published, the GPU footprint fits the serving capacity budget, prepared nodes exist, and readiness probing is configured. Each check is only applied when its query parameters are present, so the UI can preflight incrementally as a form is filled in.

get/v1/admin/inference/serving/preflight

Query parameters

publicModelIdstring

gpuModelIdstring

gpuCountPerReplicainteger

replicasDesiredinteger

weightsRefstring

nodePoolstring

The capacity lane this deployment would join, with publicModelId. Enables the rollout-peak capacity check.

rolloutRole'' | 'green'

Pass "green" to also evaluate the ROLLOUT peak: a rollout runs the old and new generations at once, so it needs the new footprint on top of what the lane already holds.

replicaRuntime'' | 'vm' | 'pod'

The runtime the deployment would use, mirroring the create request. Selects which fleet the capacity budget is evaluated against: "pod" sizes against the device-request fleet, omit (or "" or "vm") against the VFIO/VM fleet (the default).

Response

Preflight result

okboolean required

poolKeyConfiguredboolean required

OPENRELAY_POOL_API_KEY is set. Without it readiness probes return no verdicts and pool membership holds forever.