One of the most powerful features of NServiceBus is the ability to replay a message that has failed. By the time a message reaches the error queue, it will have already progressed through multiple retries via the immediate retries and delayed retries process, so you can be sure that the exception is systemic.
Often, this type of failure can be introduced by a bug that isn't found until the code is deployed. When this happens, many errors can flood into the error queue all at once. At these times, it's incredibly valuable to be able to roll back to the old version of the endpoint, and then replay the failed messages through proven code. Then you can take the time to properly troubleshoot and fix the issue before attempting a new deployment.
In this tutorial, we'll see how to use ServiceControl to monitor an NServiceBus system, and ServicePulse to replay a failed message.
To get started, download the solution, extract the archive, and then open the RetailDemo.sln file with Visual Studio 2017.
The solution is similar to the one built in the Introduction to NServiceBus tutorial, containing five projects. The ClientUI, Sales, Billing, and Shipping projects are endpoints that communicate with each other using NServiceBus messages.
The ClientUI endpoint mimics a web application and is an entry point in our system. The Sales, Billing, and Shipping endpoints contain business logic related to processing and fulfilling orders. Each endpoint references the Messages assembly, which contains the definitions of messages as POCO classes.
As shown in the diagram, the ClientUI endpoint sends a PlaceOrder command to the Sales endpoint. As a result, the Sales endpoint will publish an
OrderPlaced event using the publish/subscribe pattern, which will be received by the Billing and Shipping endpoints. Additionally, the Billing endpoint endpoint will publish an
OrderBilled endpoint that will also be received by the Shipping endpoint.
OrderBilledevent for the same order. That is a use case for a Saga and is outside of the scope of this tutorial.
The Introduction to NServiceBus tutorial on which the code is based uses the Learning Transport as its underlying message transport to move messages around. It's great for learning the NServiceBus API and for demos, but since it does not support the tools in the Particular Service Platform, we need to upgrade to a production-ready transport.
This tutorial uses the MSMQ Transport that moves messages around using Microsoft Message Queuing.
If you completed the Introduction to NServiceBus tutorial, you may want to note a few things in the code that have changed.
In the Program.cs file of every endpoint, the
LearningTransport has been replaced by the
MsmqTransport, including additional configuration that is required by MSMQ:
var transport = endpointConfiguration.UseTransport<MsmqTransport>(); endpointConfiguration.UsePersistence<InMemoryPersistence>(); endpointConfiguration.SendFailedMessagesTo("error"); endpointConfiguration.EnableInstallers();
In this snippet we can see that:
UseTransport<LearningTransport>()has been replaced by
- In-memory persistence has been selected. This is because MSMQ, unlike the learning transport, does not natively support Publish/Subscribe, so the message subscription information will be stored in memory instead.
- An error queue (
error) has been specified, which was not a required setting in the learning transport. This is important because most MSMQ systems are distributed, and a centralized error queue in the form of
error@MACHINENAMEis used for the whole system.
- EnableInstallers() is called, which instructs MSMQ to create the message queues required by the endpoints.
In addition, there is additional routing configuration in the Program.cs file in both Billing:
var routing = transport.Routing(); routing.RegisterPublisher(typeof(OrderPlaced), "Sales");
And in Shipping:
var routing = transport.Routing(); routing.RegisterPublisher(typeof(OrderPlaced), "Sales"); routing.RegisterPublisher(typeof(OrderBilled), "Billing");
This is also because MSMQ does not support native Publish/Subscribe, so subscribers need to know which logical endpoint to listen for events.
The last change in the solution is disabling delayed retries in the Sales endpoint's Program.cs file:
var recoverability = endpointConfiguration.Recoverability(); recoverability.Delayed(delayed => delayed.NumberOfRetries(0));
Since we are going to be causing a lot of messages to fail in this exercise, we'd prefer not to wait around for several rounds of delayed retries to complete.
To complete this tutorial and run the solution, we will be using the Particular Platform Installer. It will ensure that MSMQ is set up correctly, and also install two tools we need for the tutorial:
- ServiceControl is like a watchdog monitoring your system, sucking in information and making that available to other tools via a REST API. One of its functions is to monitor your error queue so that you can act on the poison messages that arrive there.
- ServicePulse is a web application aimed to be an operational dashboard for your NServiceBus system. It allows you to see failed messages, including the exception details, and provides a UI to either replay or archive failed messages.
To install the Service Platform:
- Download the Platform Installer.
- Launch the ParticularPlatform.exe you downloaded, and use it to install the Particular Service Platform according to the instructions.
Once that completes, we need to install an instance of ServiceControl. It's possible to install multiple instances of ServiceControl for different transports, so we need to configure a ServiceControl instance specifically for MSMQ.
You can launch the ServiceControl Management application in one of two ways:
- From the Start ServiceControl Management button on the last screen of the Platform Installer
- By locating ServiceControl Management in the Windows Start menu
Next, in the ServiceControl Management window, click the Add new instance button. There are a few customizations we will need to make here to configure ServiceControl.
First and foremost, under the General heading, take note of the host name and port (
localhost:33333 by default) as you will need these later.
Last, scroll down to the Queues Configuration heading:
- Change the Audit Forwarding dropdown value to Off.
Because we're just getting started with NServiceBus development, we don't need to keep copies of these messages around, so we can safely set Audit Forwarding to Off.
Now, we're ready to create and start the service:
- Click the Add button to install the ServiceControl instance as a Windows service.
- When complete, the ServiceControl Management tool will display the high-level details of the ServiceControl instance, but the instance will be in the Stopped state.
- In the upper-right corner of the ServiceControl instance details, click the Start button (the button with the Play icon) to start the service. You can also start the service from the Windows Services manager.
To check that everything is working properly, you can click on the link shown under URL, which will return a JSON response if ServiceControl is working properly. This is the API that is used to serve information to the ServicePulse and ServiceInsight tools.
Later in the exercise, we will be using ServicePulse to replay a failed message, so we should also check to make sure it is working. ServicePulse is installed as a Windows service named Particular ServicePulse and has a web-based UI, which can be accessed at
http://localhost:9090 when default settings are used. You can check to see if it is running from the Windows Services control panel. ServicePulse must be configured for the correct URL for the ServiceControl API (
localhost:33333 by default) which can be configured if a non-default ServiceControl URL is used.
The solution is configured to have multiple startup projects, so when you run the solution it should open four console applications, one for each messaging endpoint.
In the ClientUI application, press
P to place an order, and watch what happens in other windows.
It may happen too quickly to see, but the
PlaceOrder command will be sent to the Sales endpoint, which will publish events to Billing and Shipping. The Billing endpoint will then publish an event to Shipping. The entire process concludes with these two log messages displayed by the Shipping endpoint, hinting at the need for a Saga:
INFO Shipping.OrderPlacedHandler Received OrderPlaced, OrderId = 96dfd084-2bb0-46c3-b939-046e3b911102 - Should we ship now? INFO Shipping.OrderBilledHandler Received OrderBilled, OrderId = 96dfd084-2bb0-46c3-b939-046e3b911102 - Should we ship now?
Now, let's throw an exception that will make its way to the error queue. For the purposes of this exercise, we'll create a specific bug in the Sales endpoint and watch what happens when we run the endpoint.
- In the Sales endpoint, locate the PlaceOrderHandler.
- Uncomment the line that throws the exception. The code in the project contains a
#pragmadirective to prevent Visual Studio from interpreting the unreachable code after the
throwstatement as a build error.
Now, run the solution.
- In Visual Studio's Debug menu, select Detach All so that the system keeps running, but does not break into the debugger when we throw our exception.
- In the ClientUI window, place an order by pressing
In the Sales window, you will see a wall of text culminating in a red error trace. This is where NServiceBus gives up on the message and sends it to the error queue.
INFO Sales.PlaceOrderHandler Received PlaceOrder, OrderId = e927667c-b949-47ee-8ea2-f29523909784 ERROR NServiceBus.RecoverabilityExecutor Moving message '53ac6836-48ef-49dd-aabb-a67c0104a2a5' to the error queue 'error' because processing failed due to an exception: System.Exception: BOOM at < stack trace>
Because we installed ServiceControl and ServicePulse earlier, we can attempt to replay a message:
- Fix the Sales endpoint by commenting the
- Run the solution.
- Open ServicePulse at
http://localhost:9090and navigate to the Failed Messages page. Note how similar messages are grouped together for easier handling.
- Click anywhere in a message group box to see the individual failed messages in the group, including the exception message.
- Click on an individual message, and you will be able to see the stack trace for the exception, or switch tabs to see the message headers or message body.
- Click the Retry message button and watch what happens in the console windows.
- You can also back up to the Message Groups view and click Request retry to replay all the messages within that group at once. Note that ServiceControl executes message retry batches on a 30-second timer, so be patient. Eventually, the messages will be returned to their appropriate endpoints.
When the message is replayed in Sales, each endpoint picks up right where it left off. You should be able to see how useful this capability will be when failures happen in your real-life systems.
For more details see Persistence in NServiceBus.
In this tutorial, we saw how to set up the Particular Service Platform tools ServiceControl and ServicePulse to replay a failed message. With this ability, we can see the details of failed messages in ServicePulse and begin to troubleshoot what went wrong.
Perhaps the message had a previously unexpected input value which caused the bug to go undetected until the code entered production. With this knowledge in hand, we can go fix the code to validate these inputs or take some other sort of corrective action. Once the new code is deployed with the fix, we can replay the message and everything will flow through the system as if the error had never happened.
If you haven't yet, you should check out the Introduction to NServiceBus tutorial, where you'll learn how to build the solution this tutorial is based on from scratch, while learning the messaging concepts you'll need to know to build even more complex software systems with NServiceBus.