0

The N+1 trap over embeds

Queriesused in ~70% of apps

DevLog's feed renders 5 posts and the network tab shows 11 requests. Nobody wrote a bug — they wrote the obvious code: fetch the posts, then loop over them fetching each one's author. It works perfectly on your laptop with 5 posts and quietly ruins the app the day the feed has 200.

Core idea. N+1 is 1 request to list the parents plus N more requests, one per parent, to fetch something that belongs to it — a round-trip cost that scales with the size of the list, even though every individual query is fast. The fix is not 'make the queries faster'; it's asking for the related data in the SAME request via an embed, so the round-trip count stops depending on how many rows came back. The trap in the name — 'over embeds' — is that an embed can silently reintroduce the exact row-multiplication bug concept 23 already warned about, just at the JSON layer instead of the SQL layer.
Playground, this isn't a screenshot
DevLog feed · 5 posts
const posts = await supabase.from('posts').select(); for (const p of posts) { await supabase.from('profiles').select().eq('id', p.user_id); await supabase.from('comments').select('*', { count: 'exact', head: true }).eq('post_id', p.id); }
0 / 11 requests~0ms elapsed (serial round trips)
1GET posts
2GET author of "Why I switched to Postgres"
3GET author of "RLS, explained"
4GET author of "Ships on Fridays"
5GET author of "Indexing tips"
6GET author of "My Vim setup"
7GET comment count for "Why I switched to Postgres"
8GET comment count for "RLS, explained"
9GET comment count for "Ships on Fridays"
10GET comment count for "Indexing tips"
11GET comment count for "My Vim setup"
5 posts → 11 requests (1 list + 5 author lookups + 5 comment-count lookups), each paying a full round trip in series. The query underneath every single one is fast — the cost is entirely the network, multiplied by every row in the list.
↑ play a little first, then dig in