MSMQ Channel - New Version

• Download MSI file - 249 Kb
• Download source - 371 Kb

Contents

• Introduction
• Features
• Configuration
• Installation
• Usage
• Implementation
• Test
• Conclusion

Introduction

The event driven architecture model is based on the synchronous pre-processing and asynchronous post-processing design patterns. By using the Remoting infrastructure and MSMQ Technology for implementation of these patterns, it allows to encapsulate a business model from its physical deployment. This article shows the implementation and usage of the MSMQ Channel in the distributed application model such as the transactional calls, post-processing and its configuration. I also recommend to review my previous article Using MSMQ for Custom Remoting Channel for the concept and design issues.

Features

The MSMQ Channel supports the following features:

• Full Remoting custom model
• Channel Knowledge Base
• Run-time updating Channel Knowledge Base using the remoting design pattern
• Configuration file
• Public and Private queues (including the remote private queues)
• Direct format name (including the MSMQ 3.0 over Internet)
• Chaining channels mechanism
• Mapping a logical address of the connectivity to the full physical url address
• Fire&Forget transactional design pattern including the COM+/DTC transaction
• Multiple Fire&Forget calls in the transactional manner
• Passing the application info using the CallContext object (Ticket object)
• Multithreading transactional receiver with the thread-pool controller
• Transactional Retry mechanism
• Smart retry exception filter
• Send and Receive timeout administration
• Endpoint Notify mechanism to retrieve a remoting message from any queue
• Acknowledge Notification
• Exception Notification
• Support only the SAO remote object such as SingleCall and Singleton
• EventLog

Configuration

The MSMQ Channel (like any remoting channel) is a stateful object in the appDomain process identified by its unique name, for instance, msmq. Its state can be initiated during the construction time by properties located in the config file. All these properties represent a Knowledge Base (KB) of the channel. Based on their values the channel can perform a change its state. For instance, a retry mechanism needs to know how many times to re-deliver a message to the endpoint during the exception process. Some of the properties are constant and they can not be changed during their lifetime - it's called as a static KB, for instance: Channel Name. On the other hand, the dynamic properties can change a behavior of the channel on the run-time without restarting the process. Coming to the point, the MSMQ Channel has a control mechanism to :

• initiate its Knowledge Base from the config file like the standard channel
• create and modify its Knowledge Base on the runtime programmatically like another remote object

This last feature allows to build a smart distributed model with a transparent connectivity between the remote object and its consumer where its state can be dynamically changed based on the business behavior - it's called as a learning and tuning design pattern.

The following picture shows a high level design of the MSMQ Channel between the remote object and its consumer:

As you can see, each MSMQ custom channel has own Knowledge Base to hold the channel state. The KB is initiated during the registration process of the remoting channel and published as a remoting object using the channel name as its endpoint. Access to the KB is via the IKnowledgeBaseControl interface contract.

In prior of the MSMQ Channel registration, the MSMQ queues have to be available to send and receive messages. The minimum requirement is to have a destination queue for the message call (request channel). In the above picture is the ReqChannel queue. Based on the application requirements, additional queues such as retry queue and administration queue can be added. The name of the retry queue is based on the request queue name with the postfix _Retry. Note that all queues except the administration one have to be a transactional.

The receiver, based on the Knowledge Base can distributed the target response to the notification remote objects, such as the Notify Remote Object (NRO), Exception Remote Object (ERO) and Acknowledge Remote Object (ARO) using the standard remoting manner and interface contract IRemotingResponse. This feature allows to chain channels, so the response messages can be sent to the additional queues, for instance: ReqChannel_ACK, ReqChannel_ERROR, etc. using the MSMQ Sender Channel features.

The notifyurl property has a special feature for delivering the remoting IMessage to the endpoint. If its value is not empty, the IMessage is passed to the NRO notification object via its argument. This design pattern will allow to evaluate any IMessage received from the queue and make the application specific decision at the application layer such as re-routing, modifying, etc.

Message workflow

Consuming the remote object using the remoting infrastructure via the MSMQ custom channel is full transparently. However, there is necessary to know its behavior to properly setup the Knowledge Base of the Sender and Receiver channel.

Sender

The remoting message begins at the client side where the remote method is invoked on the transparent proxy. To obtain the transparent proxy of the remote object, the client calling the Activator.GetObject with the specified interface contract and the physical url address of the endpoint. To avoid explicitly (hard-coded) url address, the MSMQ Channel sender has a capability to use a logical url address related to the business connectivity model. For instance: msmq://BusinessEndpoint. This logical address is a constant and it can be hard-coded in the application layer. The MSMQ Channel scans the logical address in the Knowledge Base to find its physical interpretation. For its negative result, the Sender will suppose that this is a physical address. Note that also the hard-coded physical url address can be mapped into the another physical one. This feature allows to configure a properly deployment pattern administratively or programmatically on the fly. When this process is completed, the message is sent to the destination queue. This delivering can be under timeout control - sendtimeout limit property, special when the message has been sent over network or to disconnect able server. The message lifetime (waiting for receiver) can be also under the timeout limit specified by its receivertimeout property. When the time is elapsed, the message is stored in the administration queue (must be specified).

The Sender channel configuration part is shown in the following example:

<channel ref="msmq_sender" name="msmq" priority="1"
   updatekb="true" 
   receivetimeout="20" 
   sendtimeout="10"
   useheaders="false" 
   admin=".\private$\adminchannel" 
   lurl="Queue=.\private$\ReqChannel,
         BusinessEndpoint=.\private$\ReqChannel/endpoint,
         .\private$\ReqChannel/endpoint=.\private$\ReqChannel/endpointABC">
</channel>        

The Knowledge Base of the Sender channel:

lurl = "End = .\private$\ReqChannel/endpoint, FarEnd = 
.\private$\ReqChannel; 
tcp://localhost:9999/endpoint"

Receiver

The MSMQ Channel Receiver is more sophisticated part of the channel than its Sender side. It runs in the post-processing manner, triggered by the incoming message from the specified message queue - listener property. Its Knowledge Base has more properties to control the message workflow. The message is retrieved from the queue in the transactional manner and dispatched to the target endpoint. The result of the remote call is distributed to the notification object - ARO/ERO. There is a feature for the exception case - retry mechanism. Based on its properties such as retry counter, retrytime and retryfilter, the message can be sent for the specified time to the retry queue for its lately re-delivering to the target object. This process is repeating for a retry times and then the IMessage is sent to the specified exception object as a not deliverable message. To prevent retrying a message for a specified error (always generated it), the Knowledge Base contains a property retryfilter. The receiver scans its KB for the obtained target error and for its positive result will invoke immediately a specified exception object instead of passing it through the retry mechanism. Because the retry mechanism circulates the original message between the listener queue and its retry queue, the message lifetime can be turned off using the usetimeout property otherwise the message is sent to the administration queue when its time limit has been elapsed. Other words, delivering a message to the target object can be under the timeout limit including its retry mechanism. Note that in this case the all timeout properties for Sender and Receiver must be properly calculated.

Pulling messages from the queue is controlled by the receiver thread-pool controller to allow only a specified number of the receiver workers ran. This maxthreads property can be in the range of 0 to 20 threads, where value zero represents a turn-off listener from its incoming queue. The receiver worker represents a root thread (virtual client) to execute the IMessage. Its lifetime is depended from the target application. When all workers are busy for the notifytime time, the warning message is sent to the EventLog.

The Receiver channel configuration part is shown in the following example:

<channel ref="msmq_receiver" name="msmqRcv" 
   listener=".\private$\ReqChannel" 
   updatekb="true" 
   usetimeout="false"
   maxthreads="10" 
   retry="2" 
   retrytime="20" 
   retryfilter="No receiver registered. No connection could be 
      made because the target machine actively refused it."
   acknowledgeurl="msmq://.\private$\ReqChannel_Ack; 
      tcp://localhost:4444/endpoint" 
   exceptionurl="FormatName:DIRECT=
      http://localhost/msmq/private$/reqchannel_Error/endpoint>
</channel>

The Knowledge Base of the Receiver channel:

Note that all Url properties can have a format for chaining channels which it allows to forward messages through more than one physical channel. The following example shows this feature:

exceptionurl = "msmq://.\private$\ReqChannel_Error ; 
    tcp://localhost:4444/endpoint"

where the exceptionurl property has been specified to forward an exception to the ReqChannel_Error and then through the standard tcp channel to the endpoint object.

IKnowledgeBaseControl

The Knowledge Base is published by the IKnowledgeBaseControl interface contract and registered for remoting under the name of the channel. The following code snippet show the abstract definition of this contract:

public interface IKnowledgeBaseControl
{
   void Save(string filename); // save it into the config file
   void Load(string filename); // refresh from the config file
   void Store(string strLURL, bool bOverwrite);          
      // store logical addresses name/value, ... 
   void Store(string name, string val, bool bOverwrite); 
      // store single logical address
   void Update(string strLURL); 
     // update logical addresses name/value, ... 
   void Update(string name, string val);                 
      // update single logical address
   void RemoveAll(); // remove all logical addresses
   void Remove(string name); // remove specified logical name
   object GetAll(); // get all logical addresses
   string Get(string name); // get the specified logical name
   bool Exists(string name); // check if we have a specified logical name
   bool CanBeUpdated(); // check the status
   string Mapping(string name);   // mapping logical name by physical, 
    // if it doesn't exist, just copy 
}

Programmatically accessing the KB using the standard remoting design pattern:

string strObjectUrl = "tcp://localhost:1332/msmq";
   // channel endpoint
Type objObjectType = typeof(IKnowledgeBaseControl);
   // interface contract
IKnowledgeBaseControl ro = (IKnowledgeBaseControl)
   Activator.GetObject(objObjectType, strObjectUrl);
ro.Update("name", "value");
   // update Knowledge Base - property 'name'

IRemotingResponse

The IRemotingResponse is an interface contract between the notification object and channel. Based on the channel state, the receiver can dispatch a response to the specified notification object to perform a specific task such as recovering, re-routing, updating a business state, tracing, logging, tuning the channel's KB and etc. See the Test Sample solution - the RemotingObject.cs file for its implementation. The following code snippet described its abstract definition:

#region IRemotingResponse
[Serializable]
public class MsgSourceInfo 
{
   Acknowledgment m_Acknowledgment;
   DateTime       m_MessageSentTime;
   string         m_ResponseQueuePath;
   string         m_ReportQueuePath;

   public Acknowledgment Acknowledgment 
    { get { return m_Acknowledgment;} 
      set { m_Acknowledgment = value; }}
   public DateTime MessageSentTime 
    { get { return m_MessageSentTime;} 
      set { m_MessageSentTime = value; }}
   public string ResponseQueuePath 
    { get { return m_ResponseQueuePath;} 
      set { m_ResponseQueuePath = value; }}
   public string ReportQueuePath   { 
    get { return m_ReportQueuePath;} 
      set { m_ReportQueuePath = value; }}
}
public interface IRemotingResponse
{
   void ResponseNotify(MsgSourceInfo src, object msg);   
    // remoting notification
   void ResponseAck(object msg);                         
    // remoting response notification
   void ResponseError(object msg, Exception ex);         
    // remoting exception notification
}
#endregion

CallContext Ticket

The CallContext - Ticket object allows to pass the application specific info through the remoting infrastructure. It's an option feature of the MSMQ Channel to hide all implementation logic and simplified its usage. Any point of the remoting hierarchy including the both endpoints in the application layer such as client and remoting object can modify the Ticket. The Ticket is using the String type name/value design pattern.

#region CallContext Ticket
[Serializable]
public class LogicalTicket : NameValueCollection, 
  ILogicalThreadAffinative
{
   public LogicalTicket();                        
    // retrieve the Ticket from the current CallContext
   public LogicalTicket(params string[] args);    
    // insert name/value property into this Ticket
   public LogicalTicket(IMessage imsg);           
    // retrieve the Ticket from the IMessage object
   protected LogicalTicket(SerializationInfo info, 
    StreamingContext context); // need it for the ser/des remoting issue
   public void Set(); // set this Ticket into the current CallContext
   public void Reset(); // clear the Ticket in the current CallContext
)
#endregion

Usage of the Ticket is very generic and it can be very useful, for instance, to handle a compensation issue during the distributed transaction for the disconnected queue .

Installation

The MSMQ Channel (Sender and Receiver) requires to install the following assemblies into the GAC:

• ICustomChannel - interface contract
• MSMQChannel - Sender and Receiver channels

You can install them manually or using the .msi file included in this article. Note that before using the MSMQ Channel is necessary to have all MSMQ resources specified by the channel's KB registered and ready to send/receive messages. This task is not automated and it has to be completed manually.

The MSMQ Server/Receiver Channels can be declared globally in the machine.config file or in the host process config file. The following example shows their declaration in the channels tag:

<channels>
 <channel id="msmq_sender" type="RKiss.MSMQChannel.Sender, MSMQChannel, 
    Version=2.0.0.0, Culture=neutral, PublicKeyToken=7cfeed5d7aef9679" />
 <channel id="msmq_receiver" type="RKiss.MSMQChannel.Receiver, MSMQChannel, 
    Version=2.0.0.0, Culture=neutral, PublicKeyToken=7cfeed5d7aef9679" />
</channels>

Usage

In the previous chapters have been described the MSMQ Channel Sender and Receiver from the installation and configuration point of the view. Now, I will show you a simple and advanced usage of the MSMQ Channel in the event driven architecture. Before the first example here are some useful comments:

• Transaction context doesn't flow through the standard Remoting channels such as tcp and http
• MSMQ Channel Sender can handle itself a single transactional message
• MSMQ Channel Sender can be part of the distributed transaction (DTC)
• MSMQ Channel supports only Fire&Forget call
• Multiple Fire&Forget Calls in the transactional manner can be completed using the ServiceConfig feature (SWC)
• MSMQ Channel Receiver can dispatch a message to the local endpoint
• MSMQ Channel Receiver can dispatch a message to the chained channels
• MSMQ Channel Receiver worker represents a virtual client
• MSMQ Channel can be hosted by any process such as IIS, Windows Service, Windows Form, dllhost, etc.
• Remoting Messages are stateful
• The event driven architecture model is based on the "push model" design pattern
• Remoting infrastructure allows to isolate a logical model from its physical deployment
• Remoting channel supports the deployment patterns

The following examples show patterns:

• Business Logic model - where object and its consumer are described by the logical connectivity. Note that this pattern is driven by the application and it can be full virtualized. For instance: mobile deployment, desktop deployment, etc.
• Deployment - where the logical connectivity is mapped to the physical environment using the deployment pattern model. This layer respects the physical environment and architecture design pattern such as synchronous processor and asynchronous post-processor. The behavior of this pattern can be changed administratively using the config files. In the case of using the MSMQ Channel channel, the Knowledge Base can be updated on the runtime.

The host process config files are part of the deployment pattern. Each end of the connectivity is hosted by own AppDomain (process) and they are connected each other using the Sender and Receiver mechanism. The host process also can hosted both ends such as Receiver and Sender. The following config snippet show example of the config file:

<configuration>
  <configSections></configSections>
  <appSettings>
   add key="DataSource" value="connection string to your database"
  </appSettings>
  <system.runtime.remoting>
    <application>
      <service>
        <wellknown mode="SingleCall" type="RemoteObject, 
             RemoteObject" objectUri="endpoint" />
      </service>
      <channels>
      <channel ref="msmq_sender" name="msmq" 
          receivetimeout="20"          
          sendtimeout="10"
          admin=".\private$\adminchannel" 
          lurl="EventABC=.\private$\ReqChannel/endpoint,
                RootObjectABC=FormatName:DIRECT=
                http://localhost/msmq/private$/reqchannel,
                BusinessLogicABC=.\private$\ReqChannel; 
                 tcp://localhost:4444/endpoint"
      </channel>
      <channel ref="msmq_sender" name="msmqEvent" 
          receivetimeout="20"          
          sendtimeout="10"
          admin=".\private$\adminchannel" 
          lurl="EventABC=.\private$\EventQueue/endpoint"            
      </channel>
      <channel ref="msmq_receiver" name="msmqRcv" 
          listener=".\private$\ReqChannel" 
          usetimeout="false"
          retry="2" 
          retrytime="20" 
          retryfilter="Requested Service not found. 
                       No connection could be made because 
                       the target machine actively refused it."
          acknowledgeurl="tcp://localhost:4444/endpoint"
          exceptionurl="tcp://localhost:4444/endpoint">
      </channel>
      <channel ref="tcp" port="4444" />
    </channels>
  </application>
  <channelSinkProviders></channelSinkProviders>
  <channels>
    <channel id="msmq_sender" type="RKiss.MSMQChannel.Sender, 
             MSMQChannel, Version=2.0.0.0, Culture=neutral, 
                  PublicKeyToken=7cfeed5d7aef9679" />
    <channel id="msmq_receiver" 
             type="RKiss.MSMQChannel.Receiver, MSMQChannel, 
             Version=2.0.0.0, Culture=neutral, 
             PublicKeyToken=7cfeed5d7aef9679" />
  </channels>
  </system.runtime.remoting>
</configuration>

Example 1. - Generating Event

This is a simple example how to yield the synchronous call and make its post-processing.

The Client calls the remoting method in the fire&forget manner. This process is synchronously and transactional with a high performance profile. The client guarantees that any time the request (inquire) call will be performed and accepted without wary about the target connectivity. The request call is a stateful message and it can be post-processed with the timeout limit. Based on the application, the remote object also can call itself (recursively call) without wary about the stack overflow. As I mentioned early, the message is a stateful object and it can be modified during its traveling. Based on this state, the process can be finished and generated the event notification using the above described scenario. That's the typical behavior of the "push model"

In the case of the distributed multiple events, the Client can use a new .Net 1.1 feature - SWC (Service Without Components). I like this feature and here is its implementation from the Test Sample code (Form1.cs):

private void buttonAll_Click(object sender, System.EventArgs e)
{
   LogicalTicket ticket = null;

   try 
   {
      // CallContext
      string[] info = comboBoxTicket.Text.Split(',');
      ticket = new LogicalTicket(info);
      ticket.Set();

      // serviceconfig (.Net 1.1 feature only)
      ServiceConfig sc = new ServiceConfig();
      sc.Transaction = TransactionOption.RequiresNew;
      sc.TransactionTimeout = 20;
      sc.TrackingAppName = "MSMQChannel Test";
      sc.TransactionDescription = "buttonAll_Click"; 
      sc.TrackingEnabled = true;

      // create a transactional context
      ServiceDomain.Enter(sc);

      // all methods
      DateTime dt;
      long before = DateTime.Now.Ticks;
      string strObjectUrl = comboBoxUrl.Text;
      Type objObjectType = typeof(IPing);
      IPing ro = (IPing)Activator.GetObject(objObjectType, 
          strObjectUrl);
      ro.OneWay(string.Format("#{0}: {1}", m_counter, 
          "First message"));
      ro.Echo(string.Format("#{0}: {1}", m_counter, 
          "Second message"), out dt);
      ro.Ping(string.Format("#{0}: {1}", m_counter, 
         "Third message"));
      long after = DateTime.Now.Ticks;

      // for test purpose only
      if(comboBoxMsg.Text.IndexOf("msgbox") >= 0)
      {
         if(MessageBox.Show("Are you sure to abort it?", "Sample", 
             MessageBoxButtons.YesNo)==DialogResult.Yes)
            throw new Exception("This call has been aborted");
      }
      if(comboBoxMsg.Text.IndexOf("throw") >= 0)
         throw new Exception("This is an exception test");

      // commit transaction
      ContextUtil.SetComplete();

      // echo
      ListBoxAdd(string.Format("#{0} done, time={1}[ms]",
           m_counter++, (after-before)/10000)); 
   }
   catch(Exception ex) 
   {
      // abort transaction
      ContextUtil.SetAbort();

      Trace.WriteLine(ex.Message);
      ListBoxAddXml(ex.Message); 
   }
   finally
   {
      // clean-up 
      ServiceDomain.Leave();

      // clean-up
      if(ticket != null) 
         ticket.Reset();
   } 
}        

The above bolded code is related to SWC feature. Notice that the code snippet is a part of the Windows Form without the derived from the ServicedComponent class. Its concept is very straightforward, based on creating the transaction context using the ServiceConfig class and passing its reference to the current appDomain. The rest is the same like in the Enterprise Services. You can test this feature pressing the button All in the ClientForm program - see Test.

For the large and more complex application model, the business control state is persisted in the database. The following example shows how can be worked two different resources such as database and MSMQ in the transactional manner:

Example 2. - Updating Business State

The MSMQ Channel supports the transactional call using the DTC transaction mechanism. The transactional Root Object can work with the managed transactional recourses such as Database and MSMQ in the atomic manner. It guarantees that all resources must be committed or roll backed all of them. In the above example, it's not important which resource is going to be handled first (that's the first phase of the commit), because when the root object is out of scope all resources are going to commit respectively make a rollback if less one of them aborted. After this we can expected a message in the MSMQ Channel. This is a perfect feature of the MSMQ Channel channel to join the transactional party in the root object.

The above solution is suitable for a corporate network. What about if the business state located over Internet.

Example 3. - Updating Business State over Internet

The current technology doesn't support a Distributed Transaction Coordinator (DTC) over Internet. The Internet is based on the loosely connectivity and it's not good idea to lock the resource for the long time. There is a some "pseudo" solution to help with this issue. The Transaction needs to be divided into the synch processing and asynch post-processing transaction and its compensator. The compensator will have a responsibility to make a "rollback resources". As the above picture shows, there are two root objects for handling the resources located behind the firewall. Using the MSMQ Channel the request call can be post-processed in the target root object. Basically there are two scenarios between the root objects:

• The Synch Call is processing locally and its message is stored in the local queue like in the previous examples. The post-processing is over Internet using the chained channel mechanism (see my article Remoting Over Internet). This solution is suitable if the remote object is available.
• Using the MSMQ 3.0 feature - Direct format name over Internet. In this case, the Fire&Forget is called over Internet and the destination queue is behind the firewall. Note that the client side needs to know in advance the full direct name of the destination queue. This solution is suitable for the disconnect able applications.

Once again, the properly solution can be selected and configured in the deployment pattern using the config files.

Example 4. - "Parallel processing"

This example shows a technique how to create a "parallel business processor" using the MSMQ Channel features. Let me assume we have an application to migrate the thousands pages of the thousands records to another database schema. The task can be accomplished using the legacy synchronous design pattern, where records are processing sequentially one by one in the main loop. Everything is fine except that the processing time is too long. The event driven architecture supported by the MSMQ Channel channel can accomplish this task faster, better manageable, trace able and the components are re-usable. Here is its solution, which it requires to divided this process into the following parts:

• Pre-processor represented by the Manager component to handle pages of the records and the control process state in the database. Its logic will invoke a business remoting method with the specific state in the loop of the entry page. The calls are synchronous using the fire&forget mechanism. After this loop the Manager will update its state in the database and commit all resources. When the Manager left the scope, the next phase is starting.
• Processor represents the actually business layer, where the remoting method is processing its business logic based on the received state. The root object is a transactional SingleCall (stateless) remoting object. The incoming messages from the MSMQ Channel are performed in the thread-pool oriented workers. Each root object will update the control state in the database. Based on this state the finally event can be generated to start the post-process.
• Post-Processor is receiving the event message (call) to finalize the application process such as aborted, done, etc.

When you look at again the above picture, you will see how the "push model" is forwarding a request message from the left-to-right side. The implementation of the above pattern is straightforward and simple, much better test able than the first one - legacy sequentially solution.

To change deployment for the root objects is very simple using the chaining channels as it is shown in the following:

where the root objects after that are hosted in the separate process for instance: Windows Service.

Smart Deployment Tool - idea

As I mentioned early, the business model encapsulated services into the deployment pattern design. This step is implemented by the remoting infrastructure and its custom channel such as MSMQ Channel. It will be nice have a visual tool to administrate the application deployment pattern statically (creating the config files) and dynamically on the fly (see MSMQChannel Knowledge Base). The following picture shows this idea:

The concept of this idea is based on the remoting infrastructure. Based on the physical environment and deployment pattern the Tool will generate a host processes, config files, etc. During the runtime, the tool can subscribe its subscription to the custom sink publisher - Remoting Probe to obtain the model behavior for its tuning process (config files, Knowledge Base, etc.)

Implementation

Implementation of the Sender and Receiver of the MSMQ Custom Channel is straightforward using the standard remoting boilerplate. The design pattern of the remoting boilerplate is very well documented and also published in the advanced books. Additionally, you can find more details in my previous article [1]. Here, I am going to describe some interesting part related with the MSMQ, only.

Sender

The core of the MSMQChannel.Sender is implemented in the AsyncProcessRequest method. Based on the Knowledge Base (properties) the message call is initialized and sent to the destination queue using the specified transaction option. The following code snippet shows that:

public void AsyncProcessRequest(IClientChannelSinkStack sinkStack, 
        IMessage msgReq, ITransportHeaders headers, Stream stream) 
{
   // scope state
   string strQueuePath = null;
   string strObjectUri = null;
   Message outMsg = null;
   MessageQueueTransaction mqtx = new MessageQueueTransaction();

   try 
   {
      #region pre-processor (mapping url address)
      // split into the queuepath and endpoint
      strObjectUri = ParseLogicalUrl(m_LogicalUri, out strQueuePath);

      // update Uri property
      msgReq.Properties[MSMQChannelProperties.ObjectUri] = strObjectUri;

      // pass TransportHeaders to the receiver
      if(m_Sender.AllowHeaders == true) 
      {
         headers["__RequestUri"] = strObjectUri;
         msgReq.Properties["__RequestHeaders"] = headers;
      }
      #endregion

      #region send a remoting message 
      // set the destination queue
      m_OutQueue.Path = strQueuePath;

      // create a message
      outMsg = new Message(msgReq, new BinaryMessageFormatter()); 

      // option: timeout to pick-up a message (receive message)
      int intTimeToBeReceived = m_Sender.TimeToBeReceived;
      if(intTimeToBeReceived > 0)
         outMsg.TimeToBeReceived = 
            TimeSpan.FromSeconds(intTimeToBeReceived); 

      // option: timeout to reach destination queue (send message) 
      int intTimeToReachQueue = m_Sender.TimeToReachQueue;
      if(intTimeToReachQueue > 0)
         outMsg.TimeToReachQueue = 
            TimeSpan.FromSeconds(intTimeToReachQueue); 

      // option: notify a negative receive on the client/server side
      if(m_Sender.AdminQueuePath != MSMQChannelDefaults.EmptyStr) 
      {
         // acknowledge type (mandatory)
         outMsg.AcknowledgeType = AcknowledgeTypes.NegativeReceive | 
               AcknowledgeTypes.NotAcknowledgeReachQueue;

         // admin queue for a time-expired messages
         outMsg.AdministrationQueue = m_Sender.AdminQueue; 
      }

     // message label
     string label = string.Format("{0}/{1}, url={2}", 
        Convert.ToString(msgReq.Properties["__TypeName"]).Split(',')[0], 
        Convert.ToString(msgReq.Properties["__MethodName"]), strObjectUri);

     // Send message based on the transaction context
     if(ContextUtil.IsInTransaction == true) 
     {
        // we are already in the transaction - 
        // automatic (DTC) transactional message
        m_OutQueue.Send(outMsg, label, 
             MessageQueueTransactionType.Automatic); 
     }
     else
     {
        // this is a single transactional message 
        mqtx.Begin();
        m_OutQueue.Send(outMsg, label, mqtx); 
        mqtx.Commit();
     }
     #endregion
   }
   catch(Exception ex) 
   {
      string strError = 
        string.Format("[{0}]MSMQClientTransportSink.AsyncProcessRequest 
        error = {1}, queuepath={2},", 
        m_Sender.ChannelName, ex.Message, strQueuePath);

      m_Sender.WriteLogMsg(strError, EventLogEntryType.Error); 
      throw new Exception(strError);
   }
   finally
   {
      #region clean-up
      if(mqtx.Status == MessageQueueTransactionStatus.Pending) 
      { 
         mqtx.Abort();
      }

      if(outMsg != null)
         outMsg.Dispose();
      #endregion
   }
}        

The above method is called by the ProcessMessage in the MSMQChannel Transport Sink. Here is one interesting thing. As I mentioned early, the Sender supports only a remoting call using the Fire&Forget design pattern for any kind of the call. Its return value (IMethodReturnMessage) is always null, see the following code snippet:

public void ProcessMessage(IMessage msgReq, 
   ITransportHeaders requestHeaders, Stream requestStream, 
    out ITransportHeaders responseHeaders, out Stream responseStream)
{ 
   IMessage msgRsp = null;

   try 
   {
      #region send a remoting message 
      AsyncProcessRequest(null, msgReq, requestHeaders, requestStream);
      #endregion

      #region this is a Fire&Forget call, so we have 
               to simulate a null return message
      object retVal = null;

      // generating a null return message
      IMethodCallMessage mcm = msgReq as IMethodCallMessage;
      MethodInfo mi = mcm.MethodBase as MethodInfo;
      if(mi.ReturnType != Type.GetType("System.Void"))
      retVal = mi.ReturnType.IsValueType ? 
              Convert.ChangeType(0, mi.ReturnType) : null; 

      // return message
      msgRsp = (IMessage)new ReturnMessage(retVal, null, 0, null, mcm);
      #endregion
   }
   catch(Exception ex) 
   {
      msgRsp = new ReturnMessage(ex, (IMethodCallMessage)msgReq);
   }
   finally
   {
      #region serialize IMessage response to return back
      // Note that the BinaryFormatter is a mandatory 
      // formatter for MSMQChannel!
      BinaryFormatter bf = new BinaryFormatter();
      MemoryStream rspstream = new MemoryStream();
      bf.Serialize(rspstream, msgRsp);
      rspstream.Position = 0;

      // returns
      responseStream = rspstream;
      responseHeaders = requestHeaders;
      #endregion
   }
}    

That's all for the Sender side.

Receiver

The other side, Receiver is more complex with the event driven design pattern . Its core implementation is based on the following methods:

• Init - with a responsibility to initialize the incoming and retry queues based on the Receiver Knowledge Base. Registering the delegate for a PeekCompleted event will force the control to its handler as an action of the incoming message. This handler is a Manager method.
• Manager with a responsibility to delegate a message to its background worker based on the Thread Pool availability and Knowledge Base (maxthreads property). The manager makes a fast delegate work and at the end it will call again the BeginPeek message for the next possible incoming message. Note that the actually work is done in its worker - MessageWorker.

#region Manager
private void Manager(object source, 
    PeekCompletedEventArgs asyncResult)
{
   // Connect to the queue.
   MessageQueue mq = source as MessageQueue;

   try 
   {
      // End the asynchronous peek operation.
      mq.EndPeek(asyncResult.AsyncResult);

      // threadpool controller
      while(true) 
      { 
         // wait for condition (workers are ready, etc.)
         if(m_EventMgr.WaitOne(NotifyTime * 1000, true) == false) 
         {
            // timeout handler
            if(MaxNumberOfWorkers > 0) 
            {
               string strWarning = string.Format(
                 "[{0}]MSMQChannel.Receiver:" + 
                 "All Message Workers are busy", ChannelName);
               WriteLogMsg(strWarning, EventLogEntryType.Warning);

               // threadpool is closed, try again
               continue;
            }
            else 
            {
               m_Listening = false;
               return;
            }
         }
         // threadpool is open
         break;
      }
      // delegate work to the worker
      MessageWorkerDelegator mwd = 
            new MessageWorkerDelegator(MessageWorker);
      mwd.BeginInvoke(source, null, null);

   }
   catch(Exception ex) 
   {
      string strErr = string.Format(
            "[{0}]MSMQChannel.Receiver:Mamager failed, error = {1}", 
            ChannelName, ex.Message);
      WriteLogMsg(strErr, EventLogEntryType.Error);
   }
   finally 
   {
      // Restart the asynchronous peek operation
      if(m_Listening == true)
         mq.BeginPeek(); 
      else
         Trace.WriteLine(string.Format(
            "[{0}]MSMQChannel.Receiver.Manager has been disconnected", 
              ChannelName)); 
   }
}
#endregion        

• MessageWorker - retrieving the message from the incoming queue in the transactional manner. After de-serializing its stream, the remoting message is passed to the MessageDispatcher. Its result can be distributed based on the Knowledge Base to the acknowledge or exception remote object.
• MessageDispatcher has a responsibility to deliver a remoting message to the local endpoint or forward it to the next channel using the chaining channels mechanism.
• MessageDispatcher_NOTIFY - this is a special feature of the channel, when the remoting message can be redirected to the specified remoting notify object. The IMessage is serialized into the byte array and passed to the endpoint.
• MessageDispatcher_ACK - to perform a remoting call to the specified notification endpoint with the input argument such as IMethodReturnMessage. This notification allows to retrieve a response from the target endpoint.
• MessageDispatcher_ERROR - to perform a remoting call to the specified notification endpoint with the input arguments such as the IMethodCallMessage and exception thrown by the target endpoint.
• MessageDispatcher_RETRY has a responsibility to move a message from the incoming parent queue to the retry queue in the transactional manner based on the Knowledge Base. This feature will allow to re-deliver a remoting message to the target endpoint after the specified timeout limit.
• RetryManager has only one responsibility - a transactional movement of the timed message from the retry queue back to the parent incoming queue for its re-delivering process. The moved message represents an original one with a delayed by the waiting queue (retry queue). As the MSMQ spec stated, the administration queue can not be transactional. This restriction leaded me to design this method based on queuing the retry messages and peeking the first one in the properly time. When the message's arrived time and plus the retrytimeout limit is less that current time, the RetryManager is going to sleep. It's wakeup time will generate a next BeginPeek event and the scenario is repeating over. In the other case, when the message timeout limit is elapsed, the message is retrieved from the queue and sent to the parent queue (incoming queue) in the transactional manner. The messages between the queues are moved sequentially one by one using only one thread from the thread-pool. Here is its implementation:

#region RetryManager
private void RetryManager(object source, 
  PeekCompletedEventArgs asyncResult)
{
   // the message is going to move using the transactional manner
   MessageQueueTransaction mqtx = null;

   // Connect to the queue.
   MessageQueue mq = (MessageQueue)source;

   try 
   { 
      // End the asynchronous peek operation.
      Message msg = mq.EndPeek(asyncResult.AsyncResult);

      // check the message lifetime
      DateTime elapsedTime = msg.ArrivedTime.AddSeconds(RetryTime);

      if(DateTime.Compare(DateTime.Now, elapsedTime) >= 0) 
      {
         // it's the time to move it
         // create destination queue
         MessageQueue InQueue = new MessageQueue(ListenerPath);
         InQueue.Formatter = new BinaryMessageFormatter(); 
         mqtx = new MessageQueueTransaction();

         // move it
         mqtx.Begin(); 
         Message msgToMove = mq.Receive(new TimeSpan(0,0,0,1));
         InQueue.Send(msgToMove, mqtx);
         mqtx.Commit();
     }
     else 
     {
        // it's the time to sleep
        TimeSpan deltaTime = elapsedTime.Subtract(DateTime.Now);

        // the sleep time
        int sleeptime = Convert.ToInt32(deltaTime.TotalMilliseconds);

        // wait for the timeout or signal such as change retrytime
       // value or shutdown process
        m_EventRetryMgr.WaitOne(sleeptime, true);
     } 
   }
   catch(Exception ex) 
   {
      if(mqtx != null && mqtx.Status == 
         MessageQueueTransactionStatus.Pending) 
         mqtx.Abort();

      string strErr = string.Format(
       "[{0}]MSMQChannel.Receiver:RetryMamager failed, error = {1}",
       ChannelName, ex.Message);
      WriteLogMsg(strErr, EventLogEntryType.Error);
   }
   finally 
   {
      // Restart the asynchronous peek operation
      if(m_RetryListening == true)
         mq.BeginPeek(); 
      else
       Trace.WriteLine(string.Format(
         "[{0}]MSMQChannel.Receiver.RetryManager has been disconnected",
        ChannelName)); 
   }
}
#endregion

KnowledgeBase

The Knowledge Base in the custom channel is implemented as a remoting object with a hash table to hold all dynamic properties. The object is published under the channel name and its lifetime has been initialized for infinite time. The channel inherit this object for its fast access. The public access is available through the IKnowledgeBaseControl interface contract. The following code snippet shows the method to update the Knowledge Base:

public virtual void Update(string name, string val)
{
   // validation
   if(AllowToUpdate == false)
      throw new Exception("The Knowledge Base is locked");
   if(KB.Contains(name) == false)
      throw new Exception(string.Format(
          "The logical name '{0}' doesn't exist", name));

   // fire event
   if(OnBeforeChangeKnowledgeBase(name, val) == false) 
      return;

   // update value in the KB
   KB[name] = val;

   // fire event
   OnChangedKnowledgeBase(name, val);
}                

As you can see, its implementation is very simple and straightforward. Its public methods has a virtual signature, which it allows to override it in the derived class

Test

I built a separate solution with the following projects to test a MSMQ Custom Channel:

• Server - console host process
• Interface - Interface contract for Remote Object
• RemoteObject - Remote object endpoint
• RootObject - Enterprise Service object
• Client - Windows Form

Before starting the Test be sure that MSMQChannel and Sample have been installed - see the Installation process. Also I would like to remind you that the following MSMQ private queues have to be available:

• ReqChannel - transactional
• ReqChannel_Retry - transactional
• ReqChannel_Ack - transactional
• ReqChannel_Error - transactional
• AdminChannel - non transactional

The .msi setup will create on your desktop MSMQChannel folder. Lunch the host processes such as Server and ClientForm located in the Sample subfolder:

The Server console program using a screen to show the response from the remoting object. The ClientForm is a little bit complicated and it requires to know more details described in the previous chapters. Basically, there are two sections:

• Channel Knowledge Base - to update a Knowledge Base of the selected channel. Note that the tcp port is selected implicitly, that's why can be opened more ClientForm programs. There is a one hidden trick. Double click on the listbox item will copy its contains into the data textBox, which can be modified and updated it back to the KB.
• Remote Object - to invoke the method with a specified input argument and CallContext Ticket on the specified remote object using the url address. There are two flat buttons: Echo+Ping and All. Pressing the first button will invoke a method on the Root Object to perform a distributed transaction of the two remoting calls such as Echo and Ping. The second one will perform three remoting calls in the transaction manner direct from the ClientForm using the SWC feature. Note that the Remote Object can be thrown exception, slept or stopped by the Message Box based on the sent text in the message input argument.

Now, let's make the test:

1. Press button GetAll
2. Check the Knowledge Base in the response listBox
3. Press button Echo
4. Check the response on the console (see Server console picture)
5. Select the text 'throw exception' on the Method comboBox
6. Press button Echo
7. After 40 seconds [2 (retry) x 20s(retrytimeout)] the ResponseError should be appeared on the listBox

The same way use the other buttons, updating Knowledge Base, changing the CallContext, selecting different url addresses and so on.

Conclusion

Using the MSMQ Technology in the Remoting infrastructure will give to your Architecture model a "third dimension" - event driven pattern. In this article, I described how to incorporate and configure the MSMQ Custom Channel and encapsulate the business model from its deployment. Behind that, you have a free MSMQ Custom Channel solution including its source code and installation file. I hope you will enjoy it.

• [1] Using MSMQ for Custom Remoting Channel