Cloudflare D1 Pricing Calculator and Batch Limits

How D1 counts rows

Rows read are rows scanned, not rows returned. SELECT * FROM users reads every row, and so does a WHERE on a column without an index, even when it returns one row. Rows written count each row an INSERT, UPDATE or DELETE changes, and schema changes can count as both. Row size does not matter: a wide row and a narrow one count the same.

Every query's result carries a meta object with rows_read and rows_written. Run your real queries through wrangler d1 execute and read those numbers instead of guessing. Queries from the dashboard and Wrangler are billed like any other.

Indexes cut reads and add a write

An index on a column you filter or join on lets D1 find the matching rows without reading the rest of the table. A write that sets an indexed column also writes to the index, and that counts as a second row. Set the indexes each write touches to the number of indexes on the columns your writes set. Cloudflare says the saved reads offset the extra write in nearly all cases.

Too many SQL variables

D1 limits how many values a single query can bind, and the limit holds for each statement inside db.batch() too. A multi-row INSERT binds rows times columns values, so it reaches the limit sooner than it looks. The error reads too many SQL variables at offset, and Drizzle has an open issue asking it to chunk inserts on its own.

There are two fixes. Chunk the rows so each INSERT stays under the limit, and send the chunks in one db.batch(), which runs them in order as one transaction and rolls back if any fails. For lookups and updates over a list of ids, bind the list as one JSON string and expand it in SQL with json_each, so the whole list takes a single parameter.

At the Free plan's limits

Past the daily read or write limit, every D1 query on the account fails until the reset at 00:00 UTC. Storage works differently: once it is full, you have to delete data before you can insert again, create tables or add indexes. Moving to Workers Paid lifts the limits, typically within minutes.

D1 or Postgres through Hyperdrive

D1 is built on SQLite. Each database runs one query at a time and has a size cap that Cloudflare cannot raise, because D1 is designed for many smaller databases, such as one per tenant. An app that keeps every tenant in one large schema may suit Postgres through Hyperdrive better. The Workers pricing calculator prices either choice beside the rest of a Worker's bill.