VORQ DocsMonitor the daemon
Wire up the health endpoint, scrape metrics, read the logs and report errors to Sentry.
Health
GET /healthz on provider.metrics_port (default 9090) answers 200 ok while the daemon is
running and holds a valid coordinator session, and 503 unhealthy otherwise: the session
cannot be renewed (the coordinator is unreachable or refuses the wallet), or the daemon is
shutting down. Use it as the liveness probe of your supervisor or orchestrator.
The daemon exits if the coordinator or its model catalog cannot be reached at startup, so run it under something that restarts it.
Per-model backend health is separate: a failing backend.health probe or a tripped breaker
withdraws that model's asks without affecting /healthz. Watch vorqd_asks_published for that.
Metrics
GET /metrics serves Prometheus metrics. Useful signals:
| Symptom | Likely cause |
|---|---|
vorqd_jobs_failed_total{reason="sla_abandon"} or {reason="deadline_wait"} rising | You claim more than the backend finishes inside the SLA. Lower capacity or the entry's limits. |
vorqd_capacity_free pinned at 0 | Fully used. You may be leaving work on the table. |
vorqd_capacity_granted below your capacity | The network grants fewer slots than you asked for. The grant grows with reputation. See Capacity. |
vorqd_asks_published drops | A backend health check is failing or a breaker tripped. |
vorqd_model_free{model} at 0 while vorqd_capacity_free is not | That model's own limits are full. |
vorqd_jobs_failed_total{reason="backend_exhausted"} rising | The backend fails every attempt. Check its logs. |
vorqd_jobs_failed_total with a settle_… or session_http_… reason | Settles are not landing. See the reason list. |
Every metric and failure reason is listed in the metrics reference.
Port 9090 has no authentication, and vorqd_bid_floor_discount_pct shows how far under your
published price you will go. Do not expose it publicly.
Logs
The daemon writes one JSON object per line to stderr, at provider.log_level (default
info). Lines about a job carry its job_id, so you can follow one job end to end:
docker logs vorqd 2>&1 | grep '"0x7c65'Alert on:
- a
claimedwith nosettled,abandoned (SLA),abandoned (backend): …orrefusing claimed job (…)after the job's SLA window: a stuck job; backend tripped after N consecutive faults;the network grants N of the M slots configured.
See the log events reference.
Error reporting
Set SENTRY_DSN to send errors to Sentry. Every error log line and every unhandled
exception becomes an event, with preceding info and warning lines attached as
breadcrumbs. SENTRY_ENVIRONMENT and SENTRY_RELEASE tag the events. Stack frames carry no
local variables, so keys never leave the process, and no trace headers are added to backend
requests. Without SENTRY_DSN nothing is sent.