VORQ DocsReference
Metrics
Every Prometheus metric vorqd exports and every failure reason it reports.
GET /metrics on provider.metrics_port serves these metrics in Prometheus text format.
GET /healthz answers 200 ok or 503 unhealthy.
| Metric | Type | Labels | Meaning |
|---|---|---|---|
vorqd_jobs_claimed_total | counter | Jobs claimed. | |
vorqd_jobs_settled_total | counter | Jobs settled. | |
vorqd_jobs_failed_total | counter | reason | Jobs that ended without a settle. See failure reasons. |
vorqd_job_duration_seconds | histogram | From the start of a job's run to its settle. | |
vorqd_backend_latency_seconds | histogram | model | Time in the backend phase, including waits and retries, for jobs the backend completed. |
vorqd_capacity_free | gauge | Slots (capacity or grant, whichever is smaller) minus jobs held. | |
vorqd_capacity_granted | gauge | Slots the network grants this provider. | |
vorqd_asks_published | gauge | Asks currently on the book. | |
vorqd_model_free | gauge | model | The free sent on the model's last poll. |
vorqd_backend_load | gauge | model | Last load reading, 0..1. Only for models with a load block. |
vorqd_bid_floor_discount_pct | gauge | model | Current discount under the published ask, in percent. 0 means claims only at the configured rates. |
vorqd_load_probe_failures_total | counter | model | Load probe scrapes that produced no reading. |
vorqd_backend_retries_total | counter | model | Backend attempts retried. |
Failure reasons
Values of the reason label on vorqd_jobs_failed_total. Except where noted, the daemon also
sends a signed fail, which refunds the client immediately.
Backend and SLA
| Reason | Meaning |
|---|---|
backend_error | The backend refused the job (a non-retryable error). |
backend_exhausted | Every attempt failed. |
deadline_wait | The next wait or attempt would end past the SLA margin. |
media_input_refused | A media request or reference did not fit what was paid for or what the backend takes. |
sla_abandon | The backend finished after the SLA margin. No fail is sent; the job is left to be reclaimed. |
internal_error | An unexpected error in the daemon. No fail is sent. |
Before running (the payload or its key)
| Reason | Meaning |
|---|---|
too_short, bad_version, commitment_mismatch, bad_container | The payload is not the container the job commits to. |
undecryptable | The payload did not decrypt. |
malformed_envelope | Wrong envelope version, no input, or a missing or invalid result key. |
owner_mismatch | The envelope names a different owner than the job. |
units_in_short | The payload is larger than the declared input units allow (jobs recovered after a restart). |
unseal_failed | The daemon's box key could not open the payload key. |
not_our_bid, no_cipher | The job is designated to another provider, or no box key is loaded. |
escrow_unavailable, escrow_unreachable | No escrow client, or the escrow did not answer. |
| Escrow refusal codes | Reported verbatim, e.g. escrow_key_lost, not_claimed, wrong_wallet, stale_issued_at. |
modality_unknown | The claimed job's modality could not be established. |
task_unresolved | A recovered job's payload could not be fetched. |
claim_expired | A recovered job's SLA window had closed. |
unservable | A recovered job is for a model this daemon no longer serves. |
Settlement
| Reason | Meaning |
|---|---|
settle_<reason> | The chain refused the settle, e.g. after the deadline. No fail is sent: the job is no longer this provider's. |
settle_rejected | The registries did not recognise the signer. |
settle_reverted | The settle was mined but reverted. |
settle_upload_invalid | The result upload answered without a content id. |
settle_result_too_large | The result exceeds the coordinator's size limit (413). |
settle_http_<status> | The coordinator refused the settle with that HTTP status. |
session_http_<status> | Sign-in failed, so the settle was never sent. |
settle_transport_error | The coordinator did not answer. |