Live connectivity and response time for the database and cache layer.

Pending job counts per queue — a persistently high depth can signal a stalled worker.