Slow query log
Logging every query that takes longer than a set threshold. The fastest route from "the database feels slow" to a concrete list of things to fix.
In MySQL it's enabled with slow_query_log and long_query_time, in PostgreSQL with log_min_duration_statement. Start high, say one second, and lower the threshold as the worst cases disappear — set it to zero straight away and you drown in logs while loading the disk for nothing. The raw log is rarely usable as is; pgBadger and pt-query-digest normalize the queries and rank them by total time. The most important insight is that the query at the top is rarely the most expensive one: ten thousand queries at fifty milliseconds cost more than one at ten seconds, which is why pg_stat_statements is a necessary complement.