How can I have an async stream return with 2 data sources

user_2698407 0
Viewed 416

I have the following function that returns the standard output data, as an async stream, that results from running a System.Diagnostics.Process. Everything currently in the method works as intended; I can call it in an await foreach() loop and I get each line of output as its generated by the external exe.

private static async IAsyncEnumerable<string> ProcessAsyncStream (
    ProcessStartInfo processStartInfo)
{
   // Ensure that process is destroyed when this method exits
   using var process = new Process() { StartInfo = processStartInfo };

   // Buffer used to pass data from event-handler back to this method
   BufferBlock<string> dataBuffer = new BufferBlock<string>();

   process.OutputDataReceived += (s, e) =>
   {
      if (e.Data is null)
      {
         dataBuffer.Complete();
      }
      else
      {
         dataBuffer.Post(e.Data);
      }
   };

   // Start process and redirect output streams
   process.Start();
   process.BeginOutputReadLine();

   // Return data line by line  
   while (await dataBuffer.OutputAvailableAsync())
      yield return dataBuffer.Receive();
}

My problem is that now I need it to return both the standard output and standard error results. I made this class to hold the data from each stream.

public class ProcessData
{
   public string Error { get; set; } = "";
   public string Output { get; set; } = "";
}

and changed ProcessAsyncStream() to look like this

private static async IAsyncEnumerable<ProcessData> ProcessAsyncStream (
    ProcessStartInfo processStartInfo)
{
   // Ensure that process is destroyed when this method exits
   using var process = new Process() { StartInfo = processStartInfo };

   // Buffer used to pass data from event-handlers back to this method
   BufferBlock<string> outputDataBuffer = new BufferBlock<string>();
   BufferBlock<string> errorDataBuffer = new BufferBlock<string>();

   
   process.OutputDataReceived += (s, e) =>
   {
      if (e.Data is null)
      {
         outputDataBuffer.Complete();
      }
      else
      {
         outputDataBuffer.Post(e.Data);
      }
   };

   process.ErrorDataReceived += (s, e) =>
   {
      if (e.Data is null)
      {
         errorDataBuffer.Complete();
      }
      else
      {
         errorDataBuffer.Post(e.Data);
      }
   };

   // Start process and redirect output streams
   process.Start();
   process.BeginOutputReadLine();

   // Return data line by line
   while (await outputDataBuffer.OutputAvailableAsync()
          || await errorDataBuffer.OutputAvailableAsync())
      yield return new ProcessData() 
      {
         Error = errorDataBuffer.Receive(), 
         Output = outputDataBuffer.Receive()
      }
}

The problem is that if either buffer completes before the other than the method hangs up because that buffer's .Receive() doesn't have any data to receive. If I change the while condition to && then I won't get all the data from the other buffer.

Any suggestions?

iasyncenumerable tpl-dataflow asynchronous c#
3 Answers

Regarding the actual problem, there is an issue with the process flow of reading the blocks. The easiest solution is to just use a single buffer with multiple producers and a single consumer combined with a message packet.

The conceptual issue that you are trying to solve with the DataFlow blocks is in the fundamental nature of events an async streams. Events are pushed, and async streams are pulled.

There are several solutions that would map them together, though I think the most elegant would be just to use an Unbounded Channel as the buffer.

Channels are more modern approach than DataFlow, have less degrees of freedom, less clunky then a BufferBlock, and very lightweight and highly optimized. Additionally, I would just pass a wrapper for the different response types.

Disregarding any other problem (conceptual or otherwise).

Given

public enum MessageType
{
   Output,
   Error
}

public class Message
{
   public MessageType MessageType { get; set; }
   public string Data { get; set; }

   public Message(string data, MessageType messageType )
   {
      Data = data;
      MessageType = messageType;
   }
}

Usage

private async IAsyncEnumerable<Message> ProcessAsyncStreamAsync(
     ProcessStartInfo processStartInfo, 
     CancellationToken cancellationToken)
{
   using var process = new Process() { StartInfo = processStartInfo };

   var ch = Channel.CreateUnbounded<Message>();
   var completeCount = 0;

   void OnReceived(string data, MessageType type)
   {
      // The Interlocked memory barrier is likely overkill here
      if (data is null && Interlocked.Increment(ref completeCount) == 2)
         ch?.Writer.Complete();
      else
         ch?.Writer.WriteAsync(new Message(data, type), cancellationToken);
   }

   process.OutputDataReceived += (_, args) => OnReceived(args.Data, MessageType.Output);
   process.ErrorDataReceived += (_, args) => OnReceived(args.Data, MessageType.Error);

   // start the process 
   // ...

   await foreach (var message in ch.Reader
           .ReadAllAsync(cancellationToken)
           .ConfigureAwait(false))
      yield return message;

   // cleanup
   // ...
}

Note : completely untested

user_16129750

Complete on exit instead.

void HandleData(object sender, DataReceivedEventArgs e)
{
    if (e.Data != null) dataBuffer.Post(e.Data);
}

process.OutputDataReceived += HandleData;
process.ErrorDataReceived += HandleData;
process.Exited += (s,e) => 
{
    process.WaitForExit();
    dataBuffer.Complete();
};
user_27915400

You could use a single buffer of ProcessData items:

var buffer = new BufferBlock<ProcessData>();

Then use a custom Complete mechanism to complete the buffer when both events have propagated a null value:

process.OutputDataReceived += (s, e) =>
{
    if (e.Data is null) Complete(1);
        else buffer.Post(new ProcessData() { Output = e.Data });
};

process.ErrorDataReceived += (s, e) =>
{
    if (e.Data is null) Complete(2);
        else buffer.Post(new ProcessData() { Error = e.Data });
};

Here is an implementation of the Complete method:

bool[] completeState = new bool[2];
void Complete(int index)
{
    bool completed;
    lock (completeState.SyncRoot)
    {
        completeState[index - 1] = true;
        completed = completeState.All(v => v);
    }
    if (completed) buffer.Complete();
}
user_111785490
Related