Skip to main content
GET
Cluster Status

授权

Authorization
string
header
必填

Bearer authentication header of the form Bearer <token>, where <token> is your auth token.

响应

200 - application/json

Successful Response

gpu
GpuSummaryOut · object | null

Cluster GPU overview.

★ The four numbers satisfy total = used + free + blocked, from one source each: the executor says how many cards exist (only it knows the infrastructure), the ledger says how many are occupied -- counting only jobs that actually hold cards -- and whatever cannot be scheduled right now goes to blocked with a reason in plain words. No page should have to explain "why does it say free when I cannot submit".

queued is the fifth number and sits outside that equation on purpose: a queued job is demand on the cluster, not a slice of it. Folding it into used is how the console once showed eight occupied cards on a box where the platform was running nothing at all.

storage
StorageSummaryOut · object | null

Platform storage overview (admin only).

  • The difference from the GPU summary matters and has to reach the page: GPU numbers are computed from the ledger and are exact when asked, while these are a periodic measurement. A training process writes to disk without telling the control plane, so the only way to know is to walk the tree. Hence measured_at / stale — the page must show when this was taken.
active_jobs
JobRowOut · object[]
active_count
integer
默认值:0
profiles
ClusterProfileOut · object[]