Find the most expensive queries in a PostgreSQL database with pg_stat_statements

Documented steps

1. Add pg_stat_statements to shared_preload_libraries in postgresql.conf and restart the server — a plain CREATE EXTENSION is not sufficient on its own.
2. In the target database run: CREATE EXTENSION pg_stat_statements;
3. Rank queries by cumulative time: SELECT calls, total_exec_time, mean_exec_time, rows, query FROM pg_stat_statements ORDER BY total_exec_time DESC LIMIT 10;
4. Interpret columns: calls = execution count, total_exec_time = cumulative ms, mean_exec_time = average ms per call, rows = total rows returned. Planning-time columns are only populated if pg_stat_statements.track_planning is enabled.
5. Statements are normalized — literal constants are replaced with $1, $2, ... so structurally identical queries are grouped regardless of parameter values.
6. Reset counters when starting a fresh measurement window: SELECT pg_stat_statements_reset(); (granular reset by userid/dbid/queryid is also supported).
7. Official docs: https://www.postgresql.org/docs/current/pgstatstatements.html