A monitor inside Bulk Runs that, after a long stretch with no progress, records the run’s current phase and the work still pending — then decides carefully whether to retry, rather than blindly re-issuing paid requests.
| Term | Watchdog |
|---|---|
| Category | APIs, Reliability and Publishing |
| Also known as | Stall Watchdog |
| Where it appears | Bulk Runs |
What it means in RankGear
The watchdog lives inside Bulk Runs, the part of RankGear that pushes many pages or keywords through the same analysis in a single job. When a run appears to stall — no new results and no phase change for an extended interval — the watchdog wakes, captures which phase the run is in and what work is still queued, and surfaces that snapshot instead of letting the job hang in silence. Its purpose is to make a stuck run visible and diagnosable, not to force it forward at any cost.
How to interpret it
Read a watchdog snapshot as a diagnosis aid, not an automatic fix. Look at the visible status, the diagnostics it records, its retry decision, and any credential test together, because that combination is what tells a transient failure apart from a broken configuration. A slow provider or a timed-out page is transient and may justify one bounded retry; a bad key, a wrong endpoint, or an authorization error is a configuration fault that should stop the run rather than trigger another attempt.
| What the watchdog sees | What it tells you | Typical response |
|---|---|---|
| Timeout or slow response, valid credentials | Transient failure | One bounded, longer retry |
| Authorization or endpoint error | Invalid configuration | Stop; do not repeat the paid call |
Example
A bulk run of 400 URLs stops advancing during the scoring phase. Once the stall threshold passes, the watchdog records “phase: scoring, 137 pending” and inspects the cause. A single page that timed out is granted one bounded, longer retry; but if the provider returns an authorization error, the watchdog halts that path instead of firing the same paid request over and over.
Important considerations
- The watchdog deliberately avoids duplicating paid requests — a stall is not a cue to re-run everything, since each provider call can consume credits.
- Credentials and remote publishing targets need care: connection tests should be read-only where possible, and secrets must stay redacted in any diagnostic output.
- A recorded stall reflects the run’s own operational progress, not the quality or ranking of any page — it is a reliability signal, never a scoring one.
Related terms
Part of the RankGear glossary · how RankGear measures · the 870 factors.