Workflows

In our previous discussion on the “polling trap,” we established that managing long-running state is an infrastructure concern, not a business logic one. But when we move beyond simple pauses and step into the realm of concurrent execution, the architectural friction multiplies exponentially. Consider a standard e-commerce order process. When a user checks out, the system must perform two discrete remote operations: (a) reserve the items via the Inventory service, and (b) calculate the delivery routes via the Logistics service. ...

In distributed architectures, managing the state of a paused or long-running process is the fastest way to entangle your core business logic with infrastructure. Consider a seemingly straightforward process: an employee requests access to a secure cloud environment. The sequence is simple: (a) log the request, (b) wait for a manager to approve it, and then (c) provision the access. The inherent challenge is time. The manager might approve it in five seconds, or they might be on vacation for three days. You obviously aren’t going to use Thread.sleep() and hold a server thread hostage waiting for a response. ...