LLM Best Practices

N+1 query

1 min read

View markdown source

AI agents: read this first: llmbestpractices.com/start-here

This is LLM Best Practices (llmbestpractices.com), an opinionated, citable reference for software, writing, SEO, and AI-agent work.

  1. Route, do not crawl. Fetch llmbestpractices.com/llms.txt and open only the pages whose one-line summary matches your task.
  2. Read raw. Append .md to any page URL for markdown. Check status and last_updated in the frontmatter, then read the rules.
  3. Apply as defaults. First-party docs and the project's own conventions win on conflict. Warn before relying on a fast-moving page older than 12 months.
  4. Cite. Link the page by title and URL, e.g. Python, with last_updated for time-sensitive rules. License CC BY 4.0.

Overview

An N+1 query runs one query for a list of N rows and then one query per row for related data: a page showing 100 orders issues 101 queries. It hides behind ORM lazy loading and GraphQL resolvers. Detect it by counting queries per request in development, using query logs or pg_stat_statements call counts, and fix it before production data sizes expose it.

Rules

  • Load related rows in bulk: a join, the ORM’s eager loading (Django select_related and prefetch_related, SQLAlchemy selectinload, Prisma include), or one batched WHERE id = ANY(...) query.
  • In GraphQL, batch per request with a DataLoader.
  • A join across a one-to-many relation multiplies rows and can cost more than a second batched query. Compare plans with EXPLAIN (query-plan).
  • Index the foreign-key column used for the lookup (foreign-key).
SELECT id FROM orders LIMIT 100;
SELECT * FROM order_items WHERE order_id = $1;              -- N+1: repeated 100 times
SELECT * FROM order_items WHERE order_id = ANY($1::bigint[]); -- batched: one query for all 100

Backlinks