Runtimes

WorkflowRuntime is responsible for the following tasks:

1. Managing workflow state
2. Persisting events
3. Recovering state from events
4. Waking up the workflow at the right moments

Each runtime is created per workflow type, and its creation usually involves implementation-specific elements. WorkflowRuntime interface allows creating a WorkflowInstance based on id and input.

trait WorkflowRuntime[F[_], Ctx <: WorkflowContext, WorkflowId, Input] {

  def createInstance(id: WorkflowId, input: Input): F[WorkflowInstance[F, WCState[Ctx]]]

}

Available Runtimes

Below is an overview of the available runtimes, with their strengths and limitations:

Runtime	Description	Good For	Bad For
PekkoRuntime	Uses Pekko Persistance and Cluster Sharding to manage the state and distribution of workload.	Scalability Sophisticated requirements	Small-scale setups Low complexity use-cases
DatabaseRuntime	Relies on database storage and locking mechanisms, without any in-memory state management.	Simplicity Low memory footprint	High throughput Frequent state access
InMemoryConcurrentRuntime	Fiber-safe, cancellation-safe. Uses cats-effect primitives (Ref, Semaphore). No persistence.	Fast execution Ephemeral workflows Concurrent access	Long-running workflows State durability
InMemorySynchronizedRuntime	Thread-safe via JVM synchronization. No cats-effect dependency. No persistence.	Simple workflows Experimentation Testing	Concurrent fiber access Production use cases

WorkflowEngine

Each WorkflowRuntime is parameterized by a WorkflowInstanceEngine, a component that proxies all interactions between the instance and its state, allowing for plugging-in custom behaviours.

Below is an example of building an engine.

val knockerUpper: KnockerUpper.Agent[IO] = ???
val registry: WorkflowRegistry.Agent[IO] = ???

val engine: WorkflowInstanceEngine[IO, MyWorkflowCtx.Ctx] = WorkflowInstanceEngine
  .builder[IO]()
  .withWakeUps(knockerUpper)
  .withRegistering(registry)
  .withGreedyEvaluation
  .withLogging
  .get

KnockerUpper

KnockerUpper is a component responsible for waking up the workflow when needed, e.g., after timer has expired or retry is due. See Wake-Ups for additional details.

WorkflowRegistry

WorkflowRegistry is a component responsible for tracking the execution status of workflow instances. See Collecting Instances for additional details.