Ryan Dahl's celld brings Durable Objects to self-hosted S3-backed infrastructure

Ryan Dahl, creator of Node.js, unveiled celld, a self-hosted implementation of Cloudflare's Durable Objects and Workers. This means the stateful serverless model that was previously tied to Cloudflare's infrastructure can now run on any S3-compatible storage and your own compute. The project is compatible with Cloudflare's JavaScript APIs, so code written for Durable Objects may run on celld with minimal changes.

Running celld requires managing your own infrastructure, including an S3 bucket and a compute instance (the example uses a DigitalOcean droplet). While Dahl claims it's much cheaper, $49 a month versus $415 for 100 resident cells, Cloudflare points out that the $415 assumes all objects are continuously active; if they slumber, the cost drops to $20.65. So the cost advantage depends on usage patterns. Additionally, you take on operational overhead: you must handle scaling, availability, and maintenance that Cloudflare would otherwise manage.

The celld model uses S3 as the storage engine, which may introduce latency compared to Cloudflare's tightly integrated storage. Also, the single-threaded execution model and SQLite per object are preserved, but the underlying infrastructure differences could affect performance. The project is new and likely not as battle-tested as Cloudflare's offering. For applications that require global low-latency access, self-hosting may not match Cloudflare's edge network. Also, the compatibility is with the JavaScript APIs, but there may be subtle differences in behavior.

The article frames celld as a 'love letter' to Durable Objects, arguing that such a good primitive should run anywhere. This resonates with developers who want to avoid vendor lock-in. However, Cloudflare's response highlights that the cost comparison is not straightforward. The Hacker News reaction shows enthusiasm for running durable objects outside a single provider. This suggests that the value of the abstraction itself is recognized, but the economic and operational trade-offs are still being debated.