A small project read 33 billion D1 rows: where the reads came from and a pre-release checklist

Key points

• The author writes that his billing page showed about 33 billion D1 rows read against the 25 billion included, adding a D1 Rows Read line of $8.22, and that the cause was not user growth but queries repeatedly scanning large ranges of data.
• He explains that D1 counts rows scanned, not rows returned, and gives a rough formula: total reads = requests × rows scanned per query × queries per request; crawler traffic on the sitemap, scheduled URL-submission queues and background statistics keep running even when no person visits. Cloudflare’s current D1 pricing page also says rows read counts rows scanned rather than rows returned.
• The four amplifiers he found: a search or results page that recomputed historical statistics on every visit; the same request querying the same data more than once; a dynamic sitemap recomputed from the database on each fetch; and an SEO submission queue that ran too often with large batches.
• His fixes: precompute summary tables, reuse data already queried in a request, serve the sitemap from a periodically refreshed snapshot, lower cron frequency and batch size, and index columns used for filtering, sorting and status; he found the D1 Metrics view and the Query Insights ranking by rows read the most useful way to find the problem, and says he will not claim an exact saving from screenshots.
• Before release he now asks: will this endpoint be crawled or triggered by a schedule, how many rows does each query scan versus return, are the right indexes in place, are there duplicate queries, can the sitemap be a snapshot, can statistics be pre-aggregated, how often does the queue run, and is the billable usage trend normal — and every cron job should state its frequency, maximum items per run, estimated rows scanned and how to pause it.

Original sources

1. Published2026-06-15

   A small project read 33 billion D1 rows: where the reads came from and a pre-release checklist

   Open original source

   Sylwair reviews how a small content site on Workers Paid used more D1 rows read than the monthly allowance, explains how D1 counts scanned rows, names the four query patterns that multiplied reads, and lists the fixes and a checklist to run before shipping.