Edge Databases Compared: Cloudflare D1/KV/Durable Objects vs DynamoDB vs Cosmos DB vs Firestore

Durable Objects are not a database — they are single-threaded, stateful JavaScript objects with transactional storage. Each Durable Object instance runs in one location and provides strong consistency guarantees for its own state. They solve coordination problems: real-time collaboration, rate limiting, leader election, WebSocket rooms, and atomic counters. Think of them as miniature single-threaded state machines, not as a general-purpose data store. There is no direct equivalent on any other cloud platform.

They solve similar problems (globally distributed key-value data) but with different trade-offs. DynamoDB Global Tables provide multi-region active-active replication with last-writer-wins conflict resolution, supporting both eventually consistent and strongly consistent reads within a region. KV provides eventually consistent reads globally with writes propagating in seconds. DynamoDB supports queries, scans, and secondary indexes; KV is pure key-value with list operations. DynamoDB is more powerful; KV is simpler and cheaper for read-heavy edge workloads.

Cloudflare KV: eventual consistency (reads may be stale for seconds after write). Cloudflare D1: strong consistency for single-region writes, read replicas are eventually consistent. Durable Objects: strong consistency within a single object. DynamoDB: eventual or strong consistency per-read (your choice). Cosmos DB: five tunable consistency levels from strong to eventual. Firestore: strong consistency for all reads. The right consistency model depends on your application — financial transactions need strong consistency; content caches tolerate eventual consistency.

Cosmos DB offers one database engine with multiple API interfaces: SQL (document), MongoDB, Cassandra, Gremlin (graph), and Table (key-value). Cloudflare uses separate purpose-built services: D1 for SQL, KV for key-value, R2 for objects, Durable Objects for coordination. Cosmos DB's advantage is a unified platform with consistent management; Cloudflare's advantage is that each service is optimized for its specific use case and priced independently. The Cloudflare approach is simpler per service but requires understanding when to use each.

D1 is generally available and used in production. However, it has constraints: individual databases have size limits (10GB on paid plans), write throughput is limited by the single primary, and the feature set is still growing (no full-text search, limited analytics capabilities). For read-heavy workloads at moderate scale (content sites, configuration stores, user preferences), D1 is production-ready. For write-heavy transactional workloads or large datasets, a traditional managed database (RDS, Cloud SQL, Azure SQL) is more appropriate.

Hyperdrive is Cloudflare's connection pooling and caching proxy for existing PostgreSQL and MySQL databases. Instead of migrating to D1, you keep your existing regional database and Hyperdrive accelerates access from Workers by: (1) maintaining persistent connection pools to eliminate per-request connection overhead, (2) caching query results at the edge for repeated reads, and (3) routing queries over Cloudflare's optimized network. It bridges the gap between edge compute (Workers) and regional databases without requiring a data migration.

Both are serverless NoSQL document databases. Firestore provides strong consistency for all reads, real-time listeners for live updates, and deep Firebase integration for mobile/web apps. DynamoDB provides tunable consistency (eventual or strong per-read), higher throughput with on-demand and provisioned capacity modes, and broader AWS integration. Firestore is simpler to start with and better for real-time apps; DynamoDB handles higher write throughput and offers more control over performance tuning.

KV: 25 million read operations/month free, 1GB storage free, maximum 25MB value size. D1: 10GB per database (paid), 25 million row reads/month free, 50K row writes/month free. Durable Objects: 1GB storage per object, unlimited objects. R2: 10GB storage free, 10 million reads/month free. These limits are generous for edge workloads but constrained compared to DynamoDB (unlimited storage/throughput), Cosmos DB (unlimited storage), or Firestore (1GB free then unlimited).

Use traditional managed databases (RDS, Cloud SQL, Azure SQL) when you need: complex relational queries with JOINs across large datasets, ACID transactions spanning multiple tables, storage exceeding 10GB, high write throughput (thousands of writes/second), full-text search, or compliance requirements mandating specific database engines (Oracle, SQL Server). Edge data services excel for: read-heavy content, configuration data, session state, user preferences, and real-time coordination — workloads where low latency matters more than query complexity.