If the alert fires, check the replica lag dashboard first. Lag above 30 seconds with normal write volume usually means a long-running query is holding a snapshot open; find it with pg_stat_activity ordered by xact_start, and confirm before you kill anything. If lag is climbing with elevated writes, it is capacity, not a stuck query, and the fix is to shed read traffic rather than restart the replica. Restarting during high lag has caused two of our four longest outages.