Cloudflare has published Streamline, a reference architecture for building long-running, continuous video processing pipelines. Instead of treating a media job as a single short-lived function, Streamline combines Cloudflare Workers and Durable Objects with a containerized media engine to keep state, coordinate work, and run processing reliably over time.
- Orchestration via Durable Objects: each pipeline instance is backed by a Durable Object, which holds authoritative state and serializes coordination, so multiple Workers invocations act on a single continuous job rather than isolated requests.
- Compute on Workers: Worker scripts handle control-plane logic such as receiving events, dispatching work, and tracking pipeline progress, connecting to the Durable Object identifier for that stream.
- Containerized media engine: the actual encoding/transcoding runs in a container, decoupling heavy media processing from the request lifecycle and allowing long-running workloads.
- Cloudflare Stream integration: processed output is delivered through Cloudflare Stream, which handles storage and playback delivery.
- Long-running pattern: the design targets continuous pipelines, not one-shot uploads, so failures and restarts can resume from persisted Durable Object state.
For developers, the significance is architectural. Video workloads are awkward fits for stateless, request-scoped functions: they are long-running, stateful, and need coordination. Streamline demonstrates a concrete pattern for that gap using primitives already available on Cloudflare's developer platform, so teams can build media pipelines without standing up and operating separate queue, scheduler, and state stores. It also illustrates how Durable Objects can serve as the consistency anchor for workflows that outlive any single invocation, a pattern applicable well beyond video. Because the media engine is containerized, teams retain flexibility over codecs and tooling while Cloudflare manages coordination and delivery. The post is positioned as a demonstration, so evaluate its trade-offs — container cold starts, state sizing, and cost at scale — against your own throughput and latency requirements before adopting it in production.