effect-workflow

Use Workflow.make from effect/unstable/workflow with a unique name, payload schema, and idempotencyKey function. Register the handler with workflow.toLayer, then execute it via workflow.execute after providing a WorkflowEngine layer such as WorkflowEngine.layerMemory.

Use DurableDeferred to pause the workflow until an external signal arrives. Define the deferred with success and error schemas, await it inside the workflow, and complete it from outside using DurableDeferred.succeed with a token generated via tokenFromExecutionId or tokenFromPayload.

WorkflowEngine.layerMemory is an in-memory engine for testing that provides no durability guarantees. ClusterWorkflowEngine.layer from effect/unstable/cluster integrates with Sharding and MessageStorage so executions, activities, and durable signals survive restarts and distribute across runners.

Wrap top-level workflow effects with workflow.withCompensation, passing a rollback function that receives the success value and the failure Cause. If the workflow fails, registered compensations run in saga style, but compensation does not apply to logic nested inside activities.

Code outside activities re-runs on every replay, so HTTP calls, random values, and timestamps must be wrapped in Activity.make. Activities persist their results, and on replay the engine returns the cached result without re-executing the effect.