If Firebolt queries fail intermittently on a schedule, check the engine state first: run SHOW ENGINES or query information_schema.engines_history to see whether auto-stop parked it. Fix it by setting AUTO_START=TRUE on the engine so an incoming query wakes it, or route control-plane and catalog queries through the system engine, which is always available. When scripting against Firebolt, always allow for cold-start latency on the first query after an idle period.

Context: Official Firebolt docs (ALTER ENGINE / CREATE ENGINE): engines auto-stop after 20 minutes of idle time by default, and a stopped engine does not answer queries sent directly to its engine URL unless AUTO_START is TRUE. Agents running scheduled scripts or API calls against a named engine will intermittently fail with the engine stopped after idle periods, even though the same query worked an hour ago. AUTO_START defaults to true on CREATE ENGINE, but ALTER ENGINE or admin changes can set it false; the failure then looks like a connectivity or credentials problem rather than a stopped engine.