Get Metrics

GET /system/metrics

The fleet’s declared metric catalogue, in the Prometheus text format.

Fleet-wide, not this replica’s. The numbers come from the shared store every process flushes into, which is the whole reason the exporter is built that way: a worker has no HTTP surface, and the metrics an operator pages on are the worker’s. An endpoint that answered from the responding process’s own counters would report a fraction of the truth and look healthy doing it.

Admin floor, like /system/queue and /system/workers — but unlike those two this answer is not scoped to the caller’s tenant, because a fleet counter has no tenant. Series carrying a workspace label therefore name every workspace, which is why this sits at the admin floor rather than being the public scrape endpoint the format invites. A scraper authenticates with a service key like any other machine caller.

503 rather than an empty body when the export is off. Zero counters and no counters render identically, and the difference is “nothing is happening” versus “nothing is measuring” — the first is information and the second is an outage in the monitoring.

Role floor

Requires admin or above. The floor is a minimum, not a grant: scope filtering still applies on top of it, and a resource in another tenant answers 404 rather than 403.

Responses

Status	Description	Body
200	Prometheus text exposition.	—
503	No metric export configured.	—

Only the success paths are declared

Any operation may also answer any status in the error catalogue, and every failure body is the platform envelope — including the 422 this document types as HTTPValidationError.

Example

curl -X GET https://your-deployment.example.com/system/metrics \
  -H "Authorization: Bearer $COBRE_TOKEN"