Welcome to OStack Knowledge Sharing Community for programmer and developer-Open, Learning and Share
Welcome To Ask or Share your Answers For Others

Categories

0 votes
298 views
in Technique[技术] by (71.8m points)

c# - Why do blocks run in this order?

This is short code sample to quickly introduce you what is my question about:

using System;
using System.Linq;
using System.Threading.Tasks;
using System.Threading.Tasks.Dataflow;

namespace DataflowTest
{
    class Program
    {
        static void Main(string[] args)
        {
            var firstBlock = new TransformBlock<int, int>(x => x, new ExecutionDataflowBlockOptions
            {
                MaxDegreeOfParallelism = 4
            });
            var secondBlock = new TransformBlock<int,string>(async x =>
            {
                if (x == 12)
                {
                    await Task.Delay(5000);
                    return $"{DateTime.Now}: Message is {x} (This is delayed message!) ";
                }

                return $"{DateTime.Now}: Message is {x}";
            }, new ExecutionDataflowBlockOptions
            {
                MaxDegreeOfParallelism = 4
            });
            var thirdBlock = new ActionBlock<string>(s => Console.WriteLine(s), new ExecutionDataflowBlockOptions
            {
                MaxDegreeOfParallelism = 4
            });

            firstBlock.LinkTo(secondBlock);
            secondBlock.LinkTo(thirdBlock);

            var populateTask = Task.Run(async () =>
            {
                foreach (var x in Enumerable.Range(1, 15))
                {
                    await firstBlock.SendAsync(x);
                }
            });

            populateTask.Wait();
            secondBlock.Completion.Wait();
        }
    }
}

The output is:

09.08.2016 15:03:08: Message is 1
09.08.2016 15:03:08: Message is 5
09.08.2016 15:03:08: Message is 6
09.08.2016 15:03:08: Message is 7
09.08.2016 15:03:08: Message is 8
09.08.2016 15:03:08: Message is 9
09.08.2016 15:03:08: Message is 10
09.08.2016 15:03:08: Message is 11
09.08.2016 15:03:08: Message is 3
09.08.2016 15:03:08: Message is 2
09.08.2016 15:03:08: Message is 4
09.08.2016 15:03:13: Message is 12 (This is delayed message!)
09.08.2016 15:03:08: Message is 15
09.08.2016 15:03:08: Message is 13
09.08.2016 15:03:08: Message is 14

Why is this order and how can I change the network to get the output below?

09.08.2016 15:03:08: Message is 1
09.08.2016 15:03:08: Message is 5
09.08.2016 15:03:08: Message is 6
09.08.2016 15:03:08: Message is 7
09.08.2016 15:03:08: Message is 8
09.08.2016 15:03:08: Message is 9
09.08.2016 15:03:08: Message is 10
09.08.2016 15:03:08: Message is 11
09.08.2016 15:03:08: Message is 3
09.08.2016 15:03:08: Message is 2
09.08.2016 15:03:08: Message is 4
09.08.2016 15:03:08: Message is 15
09.08.2016 15:03:08: Message is 13
09.08.2016 15:03:08: Message is 14
09.08.2016 15:03:13: Message is 12 (This is delayed message!)

So I am wondering why should all other blocks (or tasks here) wait for the delayed block?


UPDATE

Since you guys asked me to explain my problem more detailed I made this sample that is more closer to the real pipeline I am working on. Let's say the application downloads some data and computes hash based on returned response.

using System;
using System.Diagnostics;
using System.Linq;
using System.Net.Http;
using System.Security.Cryptography;
using System.Text;
using System.Threading.Tasks;
using System.Threading.Tasks.Dataflow;

namespace DataflowTest
{
    class Program
    {
        static void Main(string[] args)
        {
            var firstBlock = new TransformBlock<int, string>(x => x.ToString(), new ExecutionDataflowBlockOptions { MaxDegreeOfParallelism = 4 });

            var secondBlock = new TransformBlock<string, Tuple<string, string>>(async x =>
            {
                using (var httpClient = new HttpClient())
                {
                    if (x == "4") await Task.Delay(5000);

                    var result = await httpClient.GetStringAsync($"http://scooterlabs.com/echo/{x}");
                    return new Tuple<string, string>(x, result);
                }
            }, new ExecutionDataflowBlockOptions { MaxDegreeOfParallelism = 4 });

            var thirdBlock = new TransformBlock<Tuple<string, string>, Tuple<string, byte[]>>(x =>
             {
                 using (var algorithm = SHA256.Create())
                 {
                     var bytes = Encoding.UTF8.GetBytes(x.Item2);
                     var hash = algorithm.ComputeHash(bytes);

                     return new Tuple<string, byte[]>(x.Item1, hash);
                 }
             }, new ExecutionDataflowBlockOptions { MaxDegreeOfParallelism = 4 });

            var fourthBlock = new ActionBlock<Tuple<string, byte[]>>(x =>
            {
                var output = $"{DateTime.Now}: Hash for element #{x.Item1}: {GetHashAsString(x.Item2)}";

                Console.WriteLine(output);
            }, new ExecutionDataflowBlockOptions { MaxDegreeOfParallelism = 4 });

            firstBlock.LinkTo(secondBlock);
            secondBlock.LinkTo(thirdBlock);
            thirdBlock.LinkTo(fourthBlock);

            var populateTasks = Enumerable.Range(1, 10).Select(x => firstBlock.SendAsync(x));
            Task.WhenAll(populateTasks).ContinueWith(x => firstBlock.Complete()).Wait();

            fourthBlock.Completion.Wait();
        }

        private static string GetHashAsString(byte[] bytes)
        {
            var sb = new StringBuilder();
            int i;
            for (i = 0; i < bytes.Length; i++)
            {
                sb.AppendFormat("{0:X2}", bytes[i]);
                if (i % 4 == 3) sb.Append(" ");
            }

            return sb.ToString();
        }
    }
}

Let's take a look at the order of the requests:

requests

This definitely makes sense. All the requests are made as soon as possible. The slow fourth request is in end of list.

Now let us see what output we have:

09.08.2016 20:44:53: Hash for element #3: 4D0AB933 EE521204 CA784F3E 248EC698 F9E4D5F3 8F23A78F 3A00E069 29E73E32
09.08.2016 20:44:53: Hash for element #2: 4D0AB933 EE521204 CA784F3E 248EC698 F9E4D5F3 8F23A78F 3A00E069 29E73E32
09.08.2016 20:44:53: Hash for element #1: 4D0AB933 EE521204 CA784F3E 248EC698 F9E4D5F3 8F23A78F 3A00E069 29E73E32
09.08.2016 20:44:58: Hash for element #6: FC86E4F8 A83036BA 365BC7EE F9371778 59A11186 ED12A43C 3885D686 5004E6B3
09.08.2016 20:44:58: Hash for element #8: FC86E4F8 A83036BA 365BC7EE F9371778 59A11186 ED12A43C 3885D686 5004E6B3
09.08.2016 20:44:58: Hash for element #9: FC86E4F8 A83036BA 365BC7EE F9371778 59A11186 ED12A43C 3885D686 5004E6B3
09.08.2016 20:44:58: Hash for element #10: FC86E4F8 A83036BA 365BC7EE F9371778 59A11186 ED12A43C 3885D686 5004E6B3
09.08.2016 20:44:58: Hash for element #4: 44A63CBF 8E27D0DD AFE5A761 AADA4E49 AA52FE8E E3D7DC82 AFEAAF1D 72A9BC7F
09.08.2016 20:44:58: Hash for element #5: FC86E4F8 A83036BA 365BC7EE F9371778 59A11186 ED12A43C 3885D686 5004E6B3
09.08.2016 20:44:58: Hash for element #7: FC86E4F8 A83036BA 365BC7EE F9371778 59A11186 ED12A43C 3885D686 5004E6B3

You can see that all the hashes after third were computed right after fourth response came.

So based on these two facts we can say that all downloaded pages were waiting for slow fourth request to be done. It would be better to not wait for fourth request and compute hashes as soon as data is downloaded. Is there any way I can achieve this?

See Question&Answers more detail:os

与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…
Welcome To Ask or Share your Answers For Others

1 Answer

0 votes
by (71.8m points)

Okay, by reference from @SirRufo I started thinking about implementing my own TransformBlock that would fit my needs and process the incoming items without a respect to ordering. This way it won't ruin the network, establishing a gap between blocks in part of downloading and would be the elegant way to go.

So I started looking at what and how can I do that. To look into sources of TransformBlock itself seemed to be a good start point so I opened TransformBlock sources on Github and started analyzing it.

Right from in the start of the class I found this interesting thing: // If parallelism is employed, we will need to support reordering messages that complete out-of-order.

// However, a developer can override this with EnsureOrdered == false.
if (dataflowBlockOptions.SupportsParallelExecution && dataflowBlockOptions.EnsureOrdered)
{
    _reorderingBuffer = new ReorderingBuffer<TOutput>(this, (owningSource, message) => ((TransformBlock<TInput, TOutput>)owningSource)._source.AddMessage(message));
}

Looks like the exact thing that we want! Let us see this EnsureOrdered option in DataflowBlockOptions class on Github:

/// <summary>Gets or sets whether ordered processing should be enforced on a block's handling of messages.</summary>
/// <remarks>
/// By default, dataflow blocks enforce ordering on the processing of messages. This means that a
/// block like <see cref="TransformBlock{TInput, TOutput}"/> will ensure that messages are output in the same
/// order they were input, even if parallelism is employed by the block and the processing of a message N finishes 
/// after the processing of a subsequent message N+1 (the block will reorder the results to maintain the input
/// ordering prior to making those results available to a consumer).  Some blocks may allow this to be relaxed,
/// however.  Setting <see cref="EnsureOrdered"/> to false tells a block that it may relax this ordering if
/// it's able to do so.  This can be beneficial if the immediacy of a processed result being made available
/// is more important than the input-to-output ordering being maintained.
/// </remarks>
public bool EnsureOrdered
{
    get { return _ensureOrdered; }
    set { _ensureOrdered = value; }
}

It was looking really good so I instantly switched to the IDE to set it. Unfortunately, there was not any setting like this:

No EnsureOrdered

I kept searching and found this note:

4.5.25-beta-23019

Package has been renamed to System.Threading.Tasks.Dataflow

When I Googled and found this package, called System.Threading.Tasks.Dataflow! So I uninstalled Microsoft.Tpl.Dataflow package and installed System.Threading.Tasks.Dataflow by issuing:

Install-Package System.Threading.Tasks.Dataflow

And there was the EnsureOrdered option. I updated the code with setting EnsureOrdered to false:

using System;
using System.Diagnostics;
using System.Linq;
using System.Net.Http;
using System.Security.Cryptography;
using System.Text;
using System.Threading.Tasks;
using System.Threading.Tasks.Dataflow;

namespace DataflowTest
{
    class Program
    {
        static void Main(string[] args)
        {
            var options = new ExecutionDataflowBlockOptions { MaxDegreeOfParallelism = 4, EnsureOrdered = false };
            var firstBlock = new TransformBlock<int, string>(x => x.ToString(), options);

            var secondBlock = new TransformBlock<string, Tuple<string, string>>(async x =>
            {
                using (var httpClient = new HttpClient())
                {
                    if (x == "4") await Task.Delay(5000);

                    var result = await httpClient.GetStringAsync($"http://scooterlabs.com/echo/{x}");
                    return new Tuple<string, string>(x, result);
                }
            }, options);

            var thirdBlock = new TransformBlock<Tuple<string, string>, Tuple<string, byte[]>>(x =>
             {
                 using (var algorithm = SHA256.Create())
                 {
                     var bytes = Encoding.UTF8.GetBytes(x.Item2);
                     var hash = algorithm.ComputeHash(bytes);

                     return new Tuple<string, byte[]>(x.Item1, hash);
                 }
             }, options);

            var fourthBlock = new ActionBlock<Tuple<string, byte[]>>(x =>
            {
                var output = $"{DateTime.Now}: Hash for element #{x.Item1}: {GetHashAsString(x.Item2)}";

                Console.WriteLine(output);
            }, options);

            firstBlock.LinkTo(secondBlock);
            secondBlock.LinkTo(thirdBlock);
            thirdBlock.LinkTo(fourthBlock);

            var populateTasks = Enumerable.Range(1, 10).Select(x => firstBlock.SendAsync(x));
            Task.WhenAll(populateTasks).ContinueWith(x => firstBlock.Complete()).Wait();

            fourthBlock.Completion.Wait();
        }

        private static string GetHashAsString(byte[] bytes)
        {
            var sb = new StringBuilder();
            int i;
            for (i = 0; i < bytes.Length; i++)
            {
                sb.AppendFormat("{0:X2}", bytes[i]);
                if (i % 4 == 3) sb.Append(" ");
            }

            return sb.ToString();
        }
    }
}

And the result output was exactly what I want:

10.08.2016 11:03:23: Hash for element #3: 8BA8A86D F25E058E 180F7AA9 1EE996B0 8D721C84 AEE8AA19 0A3F7C44 9FFED481
10.08.2016 11:03:23: Hash for element #1: 8BA8A86D F25E058E 180F7AA9 1EE996B0 8D721C84 AEE8AA19 0A3F7C44 9FFED481
10.08.2016 11:03:23: Hash for element #2: 8BA8A86D F25E058E 180F7AA9 1EE996B0 8D721C84 AEE8AA19 0A3F7C44 9FFED481
10.08.2016 11:03:23: Hash for element #10: C8C87B26 B17A6329 3F6213CD 4F9AFE0D 4F90127B AAE49EB0 7D8C15DF 74F020B1
10.08.2016 11:03:23: Hash for element #8: C8C87B26 B17A6329 3F6213CD 4F9AFE0D 4F90127B AAE49EB0 7D8C15DF 74F020B1
10.08.2016 11:03:23: Hash for element #9: C8C87B26 B17A6329 3F6213CD 4F9AFE0D 4F90127B AAE49EB0 7D8C15DF 74F020B1
10.08.2016 11:03:23: Hash for element #5: C8C87B26 B17A6329 3F6213CD 4F9AFE0D 4F90127B AAE49EB0 7D8C15DF 74F020B1
10.08.2016 11:03:23: Hash for element #7: C8C87B26 B17A6329 3F6213CD 4F9AFE0D 4F90127B AAE49EB0 7D8C15DF 74F020B1
10.08.2016 11:03:23: Hash for element #6: C8C87B26 B17A6329 3F6213CD 4F9AFE0D 4F90127B AAE49EB0 7D8C15DF 74F020B1
10.08.2016 11:03:27: Hash for element #4: FD25E52B FCD8DE81 BD38E11B 13C20B96 09473283 F25346B2 04593B70 E4357BDA

与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…
Welcome to OStack Knowledge Sharing Community for programmer and developer-Open, Learning and Share
Click Here to Ask a Question

...