Reading EXPLAIN ANALYZE like a profiler

EXPLAIN ANALYZE output looks like noise until you read it like a profiler: find where the time goes, then ask why the planner was surprised.

EXPLAIN (ANALYZE, BUFFERS)
SELECT * FROM orders WHERE customer_id = 42 AND status = 'open';

What I look at, in order

1. Actual time of the slowest node, remembering it is per loop.
2. Rows: estimated vs actual. A 100x gap means stale or missing statistics.
3. Buffers. shared read means disk; shared hit means cache.
4. Sort and hash nodes spilling to disk (Sort Method: external merge).

Common fixes

• Run ANALYZE, or raise the statistics target on skewed columns.
• Add a composite index that matches the filter and sort order.
• Rewrite OR across columns into a UNION ALL.

Most slow queries are one wrong estimate away from a fast plan.