CLOUDFLARE’S D1 is a serverless SQLite-based database designed to scale to millions of databases. The post details how the D1 control plane has been redesigned to support platforms with vast numbers of databases, hundreds of thousands of which are active at any time. Growth metrics are given: D1 has passed 5.5 million databases, with 15x growth in live non-deleted databases and 65x growth over two years (as of 8 October 2026), and 18x growth in daily active databases in the same period. The article emphasises the architectural shift away from centralized hot paths, favouring edge-centric access and distributed control processes.
Key architectural principles include failing safely with idempotent design, avoiding single-region bottlenecks in hot paths, using Workers KV for routing metadata, and leveraging Hyperdrive to reach central Postgres with efficient connection pooling and read replicas. The system separates data plane queries (executed at the edge) from control plane operations (which may touch central components), and uses eventual consistency and queuing for background processing.
Each D1 database is backed by a SQLite Durable Object; queries can be executed via an Edge API Worker or a D1 Binding in a user’s Worker, with an Eyeball Worker mediating access to the durable object. Creation flows are designed for near-user placement, with a multi-step process including DO initialization, Postgres writes, and a best-effort KV update, safeguarded by retries and heartbeat self-correction.
For enforcement and billing at scale, D1 employs asynchronous processing: usage updates are queued, deltas are batched into per-account totals via account_database_stats and account_database_stats_changes, and a ClickHouse layer handles analytics and billing reads. Edge enforcement uses Quicksilver to suspend writes when limits are exceeded.
The conclusion ties these patterns to a broader philosophy of separating control and data planes, using the right data stores for each task, and enabling scalable, edge-heavy database platforms.