Process Manager

A Process Manager is a long-running coordinator. It listens to Events, holds state between them, and emits Commands as a flow progresses through several steps.

Where a Policy is a stateless reaction to a single Event, a Process Manager is what you reach for when the flow needs memory: wait for both Events to arrive, then emit the next Command. Start a timer when this Event happens, and emit a different Command if the timer ticks before another Event arrives.

State, Correlation, Lifecycle

A Process Manager has three things every model needs to declare. It has state – the data it tracks across the flow. It has a correlation field – the identifier that ties incoming Events to the right Process Manager instance. And it has a lifecycle: the conditions under which a new instance starts, and the conditions under which an existing instance ends.

All three are explicit, not implicit. Without a starting condition, a Process Manager has no way to come into being. Without an ending condition, it would never finish. Without a correlation field, Events from different flows would land on the same instance.

Timers

A Process Manager can also schedule timers. A timer is a future moment at which the Process Manager wants to react – if no confirmation arrives within 24 hours, send a reminder. A timer is a first-class part of the Process Manager's lifecycle: declared with what triggers it, the duration after which it ticks, and the Commands it emits.

Where Process Managers Belong

A Process Manager lives inside a Bounded Context, just like an Aggregate does. It is a consistency unit in its own right, although the consistency it enforces is temporal rather than transactional: it ensures the right Commands happen in the right order, not that two pieces of state change atomically.