Troubleshooting Correlation

Article
03/02/2012

This topic applies to Windows Workflow Foundation 4 (WF4).

Correlation is used to relate workflow service messages to each other and to the correct workflow instance, but if it is not configured correctly, messages will not be received and applications will not work correctly. This topic provides an overview of several methods for troubleshooting correlation issues, and also lists some common issues that can occur when you use correlation.

Handle the UnknownMessageReceived Event

The UnknownMessageReceived event occurs when an unknown message is received by a service, including messages that cannot be correlated to an existing instance. For self-hosted services, this event can be handled in the host application.

host.UnknownMessageReceived += delegate(object sender, UnknownMessageReceivedEventArgs e)
{
    Console.WriteLine("Unknown Message Received:");
    Console.WriteLine(e.Message);
};

For Web-hosted services, this event can be handled by deriving a class from WorkflowServiceHostFactory and overriding CreateWorkflowServiceHost.

class CustomFactory : WorkflowServiceHostFactory
{
    protected override WorkflowServiceHost CreateWorkflowServiceHost(Activity activity, Uri[] baseAddresses)
    {
        // Create the WorkflowServiceHost.
        WorkflowServiceHost host = new WorkflowServiceHost(activity, baseAddresses);

        // Handle the UnknownMessageReceived event.
        host.UnknownMessageReceived += delegate(object sender, UnknownMessageReceivedEventArgs e)
        {
            Console.WriteLine("Unknown Message Received:");
            Console.WriteLine(e.Message);
        };

        return host;
    }
}

This custom WorkflowServiceHostFactory can then be specified in the svc file for the service.

<% @ServiceHost Language="C#" Service="OrderServiceWorkflow" Factory="CustomFactory" %>

When this handler is invoked, the message can be retrieved by using the Message property of the UnknownMessageReceivedEventArgs, and will resemble the following message.

<s:Envelope xmlns:s="https://schemas.xmlsoap.org/soap/envelope/">

  <s:Header>

    <To s:mustUnderstand="1" xmlns="https://schemas.microsoft.com/ws/2005/05/addr
essing/none">https://localhost:8080/OrderService</To>

    <Action s:mustUnderstand="1" xmlns="https://schemas.microsoft.com/ws/2005/05/
addressing/none">http://tempuri.org/IService/AddItem</Action>

  </s:Header>

  <s:Body>

    <AddItem xmlns="http://tempuri.org/">

      <Item>Books</Item>

    </AddItem>

  </s:Body>

</s:Envelope>

Inspecting messages dispatched to the UnknownMessageReceived handler may provide clues about why the message did not correlate to an instance of the workflow service.

Use Tracking to Monitor the Progress of the Workflow

Tracking provides a way to monitor the progress of a workflow. By default, tracking records are emitted for workflow life-cycle events, activity life-cycle events, fault propagation, and bookmark resumption. Additionally, custom tracking records can be emitted by custom activities. When troubleshooting correlation, the activity tracking records, the bookmark resumption records, and the fault propagation records are the most useful. The activity tracking records can be used to determine the current progress of the workflow and can help identify which messaging activity is currently waiting for messages. Bookmark resumption records are useful because they indicate that a message was received by the workflow, and fault propagation records provide a record of any faults in the workflow. To enable tracking, specify the desired TrackingParticipant in the WorkflowExtensions of the WorkflowServiceHost. In the following example, the ConsoleTrackingParticipant (from the Custom Tracking sample) is configured by using the default tracking profile.

host.WorkflowExtensions.Add(new ConsoleTrackingParticipant());

A tracking participant such as the ConsoleTrackingParticipant is useful for self-hosted workflow services that have a console window. For a Web-hosted service, a tracking participant that logs the tracking information to a durable store should be used, such as the built-in EtwTrackingParticipant, or a custom tracking participant that logs the information to a file, such as the TextWriterTrackingParticpant from the Tracking Using a Text File sample.

For more information about tracking and configuring tracking for a Web-hosted workflow service, see Workflow Tracking and Tracing, Configuring Tracking for a Workflow, and the Tracking samples.

Use WCF Tracing

WCF tracing provides tracing of the flow of messages to and from a workflow service. This tracing information is useful when troubleshooting correlation issues, especially for content-based correlation. To enable tracing, specify the desired trace listeners in the system.diagnostics section of the web.config file if the workflow service is Web-hosted, or the app.config file if the workflow service is self-hosted. To include the contents of the messages in the trace file, specify true for logEntireMessage in the messageLogging element in the diagnostics section of system.serviceModel. In the following example, tracing information, including the content of the messages, is configured to be written to a file that is named service.svclog.

<?xml version="1.0" encoding="utf-8" ?>
<configuration>
  <system.diagnostics>
    <sources>
      <source name="System.ServiceModel" switchValue="Information" propagateActivity="true">
        <listeners>
          <add name="corr"/>
        </listeners>
      </source>
      <source name="System.ServiceModel.MessageLogging">
        <listeners>
          <add name="corr"/>
        </listeners>
      </source>
    </sources>

    <sharedListeners>
      <add name="corr" type="System.Diagnostics.XmlWriterTraceListener" initializeData="c:\logs\service.svclog">
      </add>
    </sharedListeners>
  </system.diagnostics>

  <system.serviceModel>
    <diagnostics>
      <messageLogging logEntireMessage="true" logMalformedMessages="false"
         logMessagesAtServiceLevel="false" logMessagesAtTransportLevel="true" maxSizeOfMessageToLog="2147483647">
      </messageLogging>
    </diagnostics>
  </system.serviceModel>
</configuration>

To view the trace information that is contained in service.svclog, the Service Trace Viewer Tool (SvcTraceViewer.exe) is used. This is especially useful when troubleshooting content-based correlation issues because you can view the message contents and see exactly what is being passed, and whether it matches the CorrelationQuery for the content-based correlation. For more information about WCF tracing, see Service Trace Viewer Tool (SvcTraceViewer.exe), Configuring Tracing, and Using Tracing to Troubleshoot Your Application.

Common Context Exchange Correlation Issues

Certain types of correlation require that a specific type of binding is used for the correlation to work correctly. Examples include request-reply correlation, which requires a two-way binding such as BasicHttpBinding, and context exchange correlation, which requires a context-based binding such as BasicHttpContextBinding. Most bindings support two-way operations so this is not a common issue for request-reply correlation, but there are only a handful of context-based bindings including BasicHttpContextBinding, WSHttpContextBinding, and NetTcpContextBinding. If one of these bindings is not used, the initial call to a workflow service will succeed, but subsequent calls will fail with the following FaultException.

There is no context attached to the incoming message for the service 

and the current operation is not marked with "CanCreateInstance = true". 

In order to communicate with this service check whether the incoming binding 

supports the context protocol and has a valid context initialized.

The context information that is used for context correlation can be returned by the SendReply to the Receive activity that initializes the context correlation when using a two-way operation, or it can be specified by the caller if the operation is one-way. If the context is not sent by the caller or returned by the workflow service, then the same FaultException described previously will be returned when a subsequent operation is invoked.

For more information, see Context Exchange Correlation.

Common Request-Reply Correlation Issues

Request-reply correlation is used with a Receive/SendReply pair to implement a two-way operation in a workflow service and with a Send/ReceiveReply pair that invokes a two-way operation in another Web service. When invoking a two-way operation in a WCF service, the service can be either a traditional imperative code-based WCF service or it can be a workflow service. To use request-reply correlation a two-way binding must be used, such as BasicHttpBinding, and the operations must be two-way.

If the workflow service has two-way operations in parallel, or overlapping Receive/SendReply or Send/ReceiveReply pairs, then the implicit correlation handle management provided by WorkflowServiceHost may not be sufficient, especially in high-stress scenarios, and messages may not be correctly routed. To prevent this issue from occurring, we recommend that you always explicitly specify a CorrelationHandle when using request-reply correlation. When using the SendAndReceiveReply and ReceiveAndSendReply templates from the Messaging section of the Toolbox in the workflow designer, a CorrelationHandle is explicitly configured by default. When building a workflow by using code, the CorrelationHandle is specified in the CorrelationInitializers of the first activity in the pair. In the following example, a Receive activity is configured with an explicit CorrelationHandle specified in the RequestReplyCorrelationInitializer.

Variable<CorrelationHandle> RRHandle = new Variable<CorrelationHandle>();

Receive StartOrder = new Receive
{
    CanCreateInstance = true,
    ServiceContractName = OrderContractName,
    OperationName = "StartOrder",
    CorrelationInitializers =
    {
        new RequestReplyCorrelationInitializer
        {
            CorrelationHandle = RRHandle
        }
    }
};

SendReply ReplyToStartOrder = new SendReply
{
    Request = StartOrder,
    Content = ... // Contains the return value, if any.
};

// Construct a workflow using StartOrder and ReplyToStartOrder.

Persistence is not permitted between a Receive/SendReply pair or a Send/ReceiveReply pair. A no-persist zone is created that lasts until both activities have completed. If an activity, such as a delay activity, is in this no-persist zone and causes the workflow to become idle, the workflow will not persist even if it the host is configured to persist workflows when they become idle. If an activity, such as a persist activity, attempts to explicitly persist in the no-persist zone, a fatal exception is thrown, the workflow aborts, and a FaultException is returned to the caller. The fatal exception message is "System.InvalidOperationException: Persist activities cannot be contained within no persistence blocks.". This exception is not returned to the caller but can be observed if tracking is enabled. The message for the FaultException returned to the caller is "The operation could not be performed because WorkflowInstance '5836145b-7da2-49d0-a052-a49162adeab6' has completed".

For more information about request-reply correlation, see Request-Reply Correlation.

Common Content Correlation Issues

Content-based correlation is used when a workflow service receives multiple messages and a piece of data in the exchanged messages identifies the desired instance. Content-based correlation uses this data in the message, such as a customer number or order ID, to route messages to the correct workflow instance. This section describes several common issues that may occur when using content-based correlation.

Ensure the Identifying Data Is Unique

The data that is used to identify the instance is hashed into a correlation key. Care must be taken to guarantee that the data that is used for correlation is unique or else collisions in the hashed key might occur and cause messages to be misrouted. For example, a correlation based only on a customer name may cause a collision because there may be multiple customers who have the same name. The colon (:) should not be used as part of the data that is used to correlate the message because it is already used to delimit the message query’s key and value to form the string that is subsequently hashed. If persistence is being used, make sure that the current identifying data has not been used by a previously persisted instance. Temporarily disabling persistence can help identify this issue. WCF tracing can be used to view the calculated correlation key and is useful for debugging this kind of issue.

Race Conditions

There is a small gap in time between the service receiving a message and the correlation actually being initialized, during which follow-up messages will be ignored. If a workflow service initializes the content-based correlation by using data passed from the client over a one-way operation, and the caller sends immediate follow-up messages, these messages will be ignored during this interval. This can be avoided by using a two-way operation to initialize the correlation, or by using a TransactedReceiveScope.

Correlation Query Issues

Correlation queries are used to specify what data in a message is used to correlate the message. This data is specified by using an XPath query. If messages to a service are not being dispatched even though everything appears to be correct, one strategy for troubleshooting is to specify a literal value that matches the value of the message data instead of an XPath query. To specify a literal value, use the string function. In the following example, a MessageQuerySet is configured to use a literal value of 11445 for the OrderId and the XPath query is commented out.

MessageQuerySet = new MessageQuerySet
{
    {
        "OrderId", 
        //new XPathMessageQuery("sm:body()/tempuri:StartOrderResponse/tempuri:OrderId")
        new XPathMessageQuery("string('11445')")
    }
}

If an XPath query is configured incorrectly such that no correlation data is retrieved, a fault is returned with the following message: "A correlation query yielded an empty result set. Please ensure correlation queries for the endpoint are correctly configured." One quick way to troubleshoot this is to replace the XPath query with a literal value as described in the previous section. This issue can occur if you use the XPath query builder in the Add Correlation Initializers or CorrelatesOn Definition dialog boxes and your workflow service uses message contracts. In the following example, a message contract class is defined.

[MessageContract]
public class AddItemMessage
{
    [MessageHeader]
    public string CartId;

    [MessageBodyMember]
    public string Item;
}

This message contract is used by a Receive activity in a workflow. The CartId in the header of the message is used to correlate the message to the correct instance. If the XPath query that retrieves the CartId is created using the correlation dialogs in the workflow designer, the following incorrect XPath query is generated.

sm:body()/xg0:AddItemMessage/xg0:CartId

This XPath query would be correct if the Receive activity used parameters for the data, but since it is using a message contract it is incorrect. The following XPath query is the correct XPath query to retrieve the CartId from the header.

sm:header()/tempuri:CartId

This can be confirmed by examining the body of the message.

<s:Envelope xmlns:s="https://schemas.xmlsoap.org/soap/envelope/">
  <s:Header>
    <Action s:mustUnderstand="1" xmlns="https://schemas.microsoft.com/ws/2005/05/addressing/none">http://tempuri.org/IService/AddItem</Action>
    <h:CartId xmlns:h="http://tempuri.org/">80c95b41-c98d-4660-a6c1-99412206e54c</h:CartId>
  </s:Header>
  <s:Body>
    <AddItemMessage xmlns="http://tempuri.org/">
      <Item>Books</Item>
    </AddItemMessage>
  </s:Body>
</s:Envelope>

The following example shows a Receive activity configured for an AddItem operation that uses the previous message contract to receive data. The XPath query is correctly configured.

<Receive CorrelatesWith="[CCHandle] OperationName="AddItem" ServiceContractName="p:IService">
  <Receive.CorrelatesOn>
    <XPathMessageQuery x:Key="key1">
      <XPathMessageQuery.Namespaces>
        <ssx:XPathMessageContextMarkup>
          <x:String x:Key="xg0">http://schemas.datacontract.org/2004/07/MessageContractWFService</x:String>
        </ssx:XPathMessageContextMarkup>
      </XPathMessageQuery.Namespaces>sm:header()/tempuri:CartId</XPathMessageQuery>
  </Receive.CorrelatesOn>
  <ReceiveMessageContent DeclaredMessageType="m:AddItemMessage">
    <p1:OutArgument x:TypeArguments="m:AddItemMessage">[AddItemMessage]</p1:OutArgument>
  </ReceiveMessageContent>
</Receive>

For more information about content-based correlation, see Content Based Correlation and the Correlated Calculator sample.

Share via