How Profiling a 60-Million-Row Query Exposed a Caching Problem, Not a Query Problem

A developer investigating a slow admin dashboard endpoint assumed a GROUP BY query over 60 million rows was the core problem, but profiling revealed it completed in just 567 milliseconds. The real issue was that the database performed a full aggregation and sort of nearly 60,000 member groups on every single request, regardless of which page was being viewed. Because the LIMIT clause only applies after the entire result set is sorted, the query cost remained identical whether loading page 1 or page 500. With three leaderboard tabs refreshed frequently by multiple admins, the repeated load threatened the same Aurora cluster serving around 900,000 mobile users. Standard fixes like indexing or cursor-based pagination were ruled out since the sort key was a computed aggregate, pointing toward caching as the appropriate architectural solution.