CLOUDFLARE has announced Cloudflare K2, a public beta offering a durable, serverless event streaming primitive built on the developer platform. K2 lets you send events to a stream that is stored as an ordered log, with consumers able to read in different patterns—split reads across multiple readers or broadcast to all subscribers. It is designed to be fully serverless, scale to large data volumes, and provide long-term retention, so downtime at the consumer side won’t result in lost data.
The system sits on top of R2 object storage to deliver durable storage and strongly consistent APIs, enabling a separation of compute and storage and allowing the handling of historical data at low cost.
The article explains how K2 builds a log on object storage by batching writes into segments and using R2’s atomic operations to preserve ordering, which introduces about one second of produce latency at the 99th percentile in the initial release. It contrasts K2 with Queues and Basin Pipelines, noting K2’s strengths for high-scale data movement and long-term retention, with batching driving higher producer latency but enabling efficient fan-out.
Getting started involves creating streams via the CLI, Wrangler, dashboard or API, producing to the stream, and configuring subscriptions for consumer groups.
The beta is available now to accounts with Workers Paid subscriptions, with limits of 10 GB storage and 30 MB/s produce per stream; pricing post‑beta will be $0.04/GB for both produced and consumed data, and $0.02/GB/month for retained data, with a roadmap including higher write parallelism, key-based ordering, push-based consumers, an express tier, and Kafka compatibility.