If you've run PostgreSQL in production long enough, you've eventually looked at pg_stat_user_tables, seen a dead_tup count in the millions, and wondered what's going on. The answer is almost always: not enough vacuuming.