Some more thoughts on random_page_cost
Summary
The post revisits the PostgreSQL random_page_cost setting, arguing that the cost model is an approximation that largely reflects memory and caching effects rather than raw I/O. It explains that historical cost estimates from older, rotational drives do not translate to modern storage, and that the planner’s memory-awareness is incomplete. The author recommends system-specific tuning guided by monitoring queries in pg_stat_statements rather than chasing a universal value.