Workflows review

Workflows brings durable, multi-step execution to Workers. Define the steps and the platform handles retries, long sleeps, and resuming after failures while persisting each result. It turns orchestration code full of queues and cron checks into a readable function.

In practice, it has been solid. Retries behave as documented, sleeps measured in days resume on time, and failed runs pick up where they stopped. The observability is now good enough to debug a misbehaving workflow. It handles provisioning, payment flows, and long-running jobs dependably.

The retry story keeps improving. The newer retry function makes it easy to extract retry headers from an upstream response and honor them, so backing off the way an API asks for takes a few lines instead of hand-rolled delay logic.

Workflows is built on Durable Objects, but packages their coordination and persistence behind something predictable. If a design is pushing you toward raw Durable Objects, try Workflows first. It loses two points because the tooling is still growing, it does not suit high-frequency, low-latency coordination, and the pricing has drifted: per-step charges now sit on top of the originally promised compute time and step state storage. This is a batch-and-steps tool, and it should stay one.