Dru Sellers

Building a Pipeline in MassTransit

937 words · 5 minute read

At my new employer we are rebuilding our order processing pipeline to provide greater insight into the various stages of an order. This is not a unique problem, so I thought I would share how I built it using the MassTransit family of tools:

Context: C# and .Net

The pipeline

  • Receive partner command
    • Convert to standard model
  • Store order data in primary system
  • Initiate Order Flow w/ Courier Routing Slip
    • allocate order
    • approve order
    • create shipment
    • enter into Warehouse Management System
  • monitor order (and update for customers to see) for
    • in wave (picking)
    • packing order
    • ship order
  • close order out
    • close shipment
    • close order

Accepting commands at the edge

At the edge is our company HTTP API, here we accept incoming commands from our integration partners and convert them into our internal model.

public class SalesOrderEndpoint
  MassTransit.IServiceBus _bus;

  public HttpResponse Post(HttpRequest request)
    var command = ConvertToRegisterSalesOrderCommand(request);
    var endpointAddress = GetPriorityBasedEndpoint(command);

Nothing too surprising here I hope. We simply convert it from the integration partner model into our internal model. Then we get a priority appropriate input queue, this lets us maintain a high throughput for all customers, and avoids letting any one customer take up the whole queue, starving other customers.

Tracking IDs

Much like when I order something from Amazon, I want to be able to track my order through my pipeline. In order to do so I need to identify and create a few numbers. The first is the ID that I am going to use for idempotency - in this case I will use RegisterSalesOrder.ReferenceId - this maps to the external systems unique id. I additionally want to be able to track the progress of the unique order through the pipeline each time (if it fails for instance) so I also give it an instance id, which will be a Guid so I can create it in the edge code.

Idempotency check

I often get asked how I do idempotency in my messages. Well here at the beginning of the pipeline is the best way i think.

public class RegisterSalesOrderConsumer :
  public void Consume(IConsumeContext<RegisterSalesOrder> context)
      _log.Info("Ignoring Order, already processed");


Getting everyone to play nice

If you haven't seen MassTransit.Courier yet, you should take a second to read Chris' blog on it. We are going to use Courier to orchestrate the various steps in our pipeline.

Here we define some sample activities similar to what we might see in real code.

public class RegisterSalesOrderActivity :
  Activity<RegisterSalesOrder, UndoSalesOrder>
  public ExecutionResult Execute(Execution<RegisterSalesOrder> execution)
    var order = execution.Arguments;

    var id = StoreOrder(order);
    return execution.Completed(new UndoSalesOrder{ OrderId = id });

  public CompensationResult Compensate(Compensation<UndoSalesOrder> compensation)
    var log = compensation.Log;

public class AllocateOrderActivity :
  Activity<AllocateOrder, DeallocateOrder>
  public ExecutionResult Execute(Execution<AllocateOrder> execution)
    var order = execution.Arguments;

    var allocationId = AllocateOrder(order);

    return execution.Completed(new DeallocateOrder{ AllocationId = allocationId });

  public CompensationResult Compensate(Compensation<DeallocateOrder> compensation)
    var log = compensation.Log;

Now that we have the Activities that we want to orchestrate, we can build a routing slip that will be executed on the bus.

public class BuildSlipAndExecute
  IServiceBus _bus;

  //init in ctor

  public void Execute()
    var routingSlip = //build it up

So, we use routing slips and activities from corourier to automate the various steps in our process and ensure that they get completed.

  • what if there is a problem
    • compensatable
    • non-compensatable
  • what if it stops half way into the routing slip?

State Machines

public class OrderState : StateMachineInstance
  public State CurrentState { get; set; }
  public DateTime ReceivedOn { get; set; }
  public DateTime? ShippedOn { get; set; }
  public string TrackingNumber { get; set; }
public class OrderWorkflow :
    public OrderWorkflow()
      Event(() => Received);
      Event(() => Allocated);
      Event(() => Picked);
      Event(() => Shipped);
      State(() => AllocatingOrder);
      State(() => PickingOrder);
      State(() => PackingOrder);
      State(() => PendingShipment);
      State(() => Complete);
          .Then((instance,data) =>
            instance.TrackingNumber = data.TrackingNumber;
            instance.ShippedOn = DateTime.Now;

  public State AllocatingOrder { get; set; }
  public State PickingOrder { get; set; }
  public State PackingOrder { get; set; }
  public State PendingShipment { get; set; }
  public State Complete { get; set; }
  public Event<SalesOrderReceived> Received { get; set; }
  public Event<SalesOrderAllocated> Allocated { get; set; }
  public Event<SalesOrderPicked> Picked { get; set; }
  public Event<SalesOrderShipped> Shipped { get; set; }

Because we can persist Automatonymous state machines to a database with their state, we can easily query these guys back out to see where our 'orders' are.

  • external saga for the customer
  • that is 'fed?' by internal sagas

how do statemachines get subscribed

Human Response Monitors?

automated steps are orchestrated by 'routing slips' (routing slip entry/creation points) then something has to watch the humans, and start the next routing slip

the saga subscribes to the public events

  • automatonymous

internal events / routing slip completed

  • routing slip correlation table
    • watches for the routing slip completed events
      • faulted / failed / completed

3 different routing slip places

  • routing slip completed is a global event
    • trying to decide if it matters or not