Quota diagnostics

stado quota show and stado quota show --json read live provider quota, the reservation overlay, and the running machine count before reporting available accelerators. Stado Desktop's Fleet → Compute providers screen offers Read quota and reservations per provider through the same native operation and keeps its refusal in the command receipt.

If GCP returns an instance without a name or lifecycle status, an unknown status, or a malformed accelerator type or count, Stado refuses the quota summary instead of treating that instance as consuming no GPU. The refusal names the GCP instance and zone where available, the missing or malformed field, and why running accelerator use cannot be counted. A count that would overflow is refused too. Inspect the instance through GCP and correct the upstream resource or provider response; do not interpret a failed summary as free capacity. A GCP instance with no guestAccelerators field has no accelerators to count.

The GCE aggregated-list read also refuses a malformed page, an unreachable scope, a malformed zone or instance list, or an unrecognised warning rather than silently dropping that part of the fleet. Google's explicit NO_RESULTS_ON_PAGE warning is the empty-result case. The refusal names the aggregated-list operation, page path and failing scope or field.

stado quota show --json
stado resources show --provider gcp --json

The first command is the quota answer or a refusal, not an empty successful report. The second reads the selected provider's live machines without publishing capacity records. See Stado Desktop for the graphical receipt.

Quota operations

stado quota inspects GPU quota and submits increase requests for enabled providers in WC_PROVIDERS order. show is the default: live quota minus reservation minus running machines. A failed provider quota fetch remains visible rather than being read as zero quota or an empty fleet.

GCP uses the deployment's declared GCP_PROJECT or config.json project. Without either, show, catalog, request create and scheduling refuse no GCP project is declared: set GCP_PROJECT before credentials are read. A pending request with no Google-reported preferred value says not stated.

stado quota show
stado quota catalog
stado quota request create <ACCELERATOR> --to <CAPACITY> --justification "<why this capacity is needed>"
stado quota request create --every-family --to <CAPACITY> --justification "<why>"
stado quota request list
stado quota ticket reply --provider azure --dry-run
stado quota ticket escalate --provider azure --dry-run

request create submits a Quotas API request per provider and region for one accelerator or every catalog family, but not both. Regions default to the dispatch regions; GCP needs $WC_QUOTA_CONTACT_EMAIL or --email, and --justification is mandatory. Stado invents no reviewer explanation. request list shows in-flight requests and support communications. ticket reply and ticket escalate answer Azure support tickets; --dry-run posts nothing. They refuse a malformed open-ticket list, a missing ticket name or latest-message sender, or a missing subscription id and, for escalation, quota id, quoting the upstream answer. A message without a body is an error row; missing title region or message fields are reported as unstated. Both print one table, TICKET REGION OK ACTION, and a count of the rows the provider said succeeded; a field a row does not carry, its OK included, reads not reported and is not counted, and a dry-run row says dry_run → would be replied or would be escalated. Stado Desktop asks for confirmation before a real request or ticket reply. See jobs for dispatch and costs for spend.

Source: this website