How do I collect return values from Parallel.ForEach?

Question

I'm calling a slow webservice in parallel. Things were great until I realized I need to get some information back from the service. But I don't see where to get the values back. I can't write to the database, HttpContext.Current appears to be null inside of a method called using Parallel.ForEach

Below is a sample program (in your mind, please imagine a slow web service instead of a string concatenation)

using System;
using System.Threading.Tasks;

class Program
{
    static void Main(string[] args)
    {
        WordMaker m = new WordMaker();
        m.MakeIt();
    }
    public class WordMaker
    {
        public void MakeIt()
        {
            string[] words = { "ack", "ook" };
            ParallelLoopResult result = Parallel.ForEach(words, word => AddB(word));
            Console.WriteLine("Where did my results go?");
            Console.ReadKey();
        }
        public string AddB(string word)
        {
            return "b" + word;
        }
    }

}

A different overload of Parallel.ForEach may be what you want: msdn.microsoft.com/en-us/library/dd991486.aspx — Austin Salonen
– Austin Salonen, Commented Sep 26, 2012 at 21:45
Unfortunately that's not really something you can do like that. Parallel.Foreach() just wasn't built for keeping track of returns. I would, however, suggest using ref parameters in your AddB function. That might do it. — Phillip Schmidt
– Phillip Schmidt, Commented Sep 26, 2012 at 21:46
@PhillipSchmidt: Not with the overload used in the example anyway... — Austin Salonen
– Austin Salonen, Commented Sep 26, 2012 at 21:47
@AustinSalonen What's another overload that would work? I'm not arguing, I just wasn't aware of anything that would do what he wants. — Phillip Schmidt
– Phillip Schmidt, Commented Sep 26, 2012 at 21:49
@AustinSalonen you expect me to use my eyes before asking stupid questions? Jeez. Just kidding - reading it now. — Phillip Schmidt
– Phillip Schmidt, Commented Sep 26, 2012 at 21:55

Theodor Zoulias · Accepted Answer · 2022-01-26 07:01:30Z

86

You've discarded it in here.

ParallelLoopResult result = Parallel.ForEach(words, word => AddB(word));

You probably want something like,

ParallelLoopResult result = Parallel.ForEach(words, word =>
{
    string result = AddB(word);
    // do something with result
});

If you want some sort of collection at the end of this, consider using one of the collections under System.Collections.Concurrent, like ConcurrentBag

ConcurrentBag<string> resultCollection = new ConcurrentBag<string>();
ParallelLoopResult result = Parallel.ForEach(words, word =>
{
    resultCollection.Add(AddB(word));
});

// Do something with the result

edited Jan 26, 2022 at 7:01

Theodor Zoulias

46.1k8 gold badges112 silver badges155 bronze badges

answered Sep 26, 2012 at 21:42

M Afifi

4,8152 gold badges31 silver badges48 bronze badges

Sign up to request clarification or add additional context in comments.

10 Comments

usr Over a year ago

I think the ParallelLoopResult does nothing useful here. +1 though

Jorge Aguirre Over a year ago

.AsParallel() in LINQ would be a lot better

Russell Horwood Over a year ago

@eranotzap That's the issue that is solved by using a ConcurrentBag instead of a List.

eran otzap Over a year ago

@NineTails so it is an issue. only adding not adding and removing. Is only adding to a list a concurrency issue. maybe it's done in an atomic way. meaning that the index in the backing array might be incremented atomically

Theodor Zoulias Over a year ago

@PatrickfromNDependteam your latest edit (revision 6) deviates a lot from the original answer, to a point that invalidates the voting system of the site. I would suggest to post your contribution as a separate answer instead.

|

Steves · Accepted Answer · 2016-09-28 08:11:01Z

47

Your may consider using AsParallel extension method of IEnumerable, it will take care of the concurrency for you and collect the results.

words.AsParallel().Select(AddB).ToArray()

Synchronisation (e.g. locks or concurrent collections that use locks) are usually bottleneck of concurrent algorithms. The best is to avoid synchronisation as much as possible. I am guessing that AsParallel uses something smarter like putting all the items produced on single thread into a local non-concurrent collection and then combining these at the end.

answered Sep 28, 2016 at 8:11

Steves

3,27023 silver badges25 bronze badges

6 Comments

Jorge Aguirre Over a year ago

This is significantly better.

Joy George Kunjikkuru Over a year ago

When I try this the code in select doesn't seems running in different threads. But Parallel.Foreach does it.

Steves Over a year ago

Are you sure you're using AsParallel? What makes you think it's not using more threads?

Theodor Zoulias Over a year ago

Optionally you can also add .AsOrdered() after the .AsParallel(), to receive the results in the order of the input values.

Dominik Over a year ago

Is there a way to do this, for a method AddB that takes 3 arguments, of which only the first is the IEnumerable?

|

Minh Nguyen · Accepted Answer · 2021-02-01 17:50:12Z

15

Do not use ConcurrentBag to collect results as it is slower. Use local lock instead.

var resultCollection = new List<string>();
object localLockObject = new object();

Parallel.ForEach<string, List<string>>(
      words,
      () => { return new List<string>(); },
      (word, state, localList) =>
      {
         localList.Add(AddB(word));
         return localList;
      },
      (finalResult) => { lock (localLockObject) resultCollection.AddRange(finalResult); }
); 

// Do something with resultCollection here

edited Feb 1, 2021 at 17:50

answered Nov 25, 2014 at 13:21

Minh Nguyen

2,2112 gold badges31 silver badges34 bronze badges

3 Comments

dinesh ygv Over a year ago

Do you have any stats to show that ConcurrentBag is slower than using our own object lock? I just want to know how slow it is, since it makes my code look cleaner than using object lock.

Matas Vaitkevicius Over a year ago

@dineshygv IMHO Difference is negligible stackoverflow.com/questions/2950955/…

Steves Over a year ago

Or don't use any locking at all ;-)

Overlord Zurg · Accepted Answer · 2017-01-20 15:20:54Z

8

This seems safe, fast, and simple:

    public string[] MakeIt() {
        string[] words = { "ack", "ook" };
        string[] results = new string[words.Length];
        ParallelLoopResult result =
            Parallel.For(0, words.Length, i => results[i] = AddB(words[i]));
        return results;
    }

answered Jan 20, 2017 at 15:20

Overlord Zurg

3,8322 gold badges26 silver badges32 bronze badges

2 Comments

Steves Over a year ago

This is likely to cause cache ping-pong, although still substantially better than concurrent collection.

Oguz Karadenizli Over a year ago

Very very wise approach! @overlord-zurg

Patrick from NDepend team · Accepted Answer · 2022-01-26 07:55:23Z

6

In the particular cases where the size of the collection is know upfront - which is often the case in practice - an array can be use instead of an expensive concurrent collection. There is no risk of collision since each loop accesses its own slot in the ouputs array. As a bonus outputs are stored with the same order as inputs:

const int NB_WORDS = 1000;
var inputs = new string[NB_WORDS];
for(var i= 0; i < NB_WORDS; i++) { inputs[i] = i.ToString(); }

var outputs = new string[NB_WORDS];

Parallel.For(0, NB_WORDS, index => {
   string word = inputs[index];
   string result = word + word; // Operation on word
   outputs[index] = result; // No need of a concurrent collection to store the result!
});

Debug.Assert(outputs.All(result => !string.IsNullOrEmpty(result)));

edited Jan 26, 2022 at 7:55

answered Jan 26, 2022 at 7:06

Patrick from NDepend team

14.1k6 gold badges71 silver badges111 bronze badges

2 Comments

Theodor Zoulias Over a year ago

Instead of enumerating an Enumerable.Range with the Parallel.ForEach, isn't it simpler to use the Parallel.For method?

Patrick from NDepend team Over a year ago

@TheodorZoulias sure thanks

MichaC · Accepted Answer · 2015-04-04 00:58:57Z

3

How about something like this:

public class WordContainer
{
    public WordContainer(string word)
    {
        Word = word;
    }

    public string Word { get; private set; }
    public string Result { get; set; }
}

public class WordMaker
{
    public void MakeIt()
    {
        string[] words = { "ack", "ook" };
        List<WordContainer> containers = words.Select(w => new WordContainer(w)).ToList();

        Parallel.ForEach(containers, AddB);

        //containers.ForEach(c => Console.WriteLine(c.Result));
        foreach (var container in containers)
        {
            Console.WriteLine(container.Result);
        }

        Console.ReadKey();
    }

    public void AddB(WordContainer container)
    {
        container.Result = "b" + container.Word;
    }
}

I believe the locking or concurrent objects isn't necessary unless you need the results to interact with one another (like you were computing a sum or combining all the words). In this case ForEach neatly breaks your original list up and hands each thread its own object that it can manipulate all it wants without worrying about interfering with the other threads.

edited Apr 4, 2015 at 0:58

answered Apr 2, 2015 at 23:36

MichaC

2,8442 gold badges20 silver badges20 bronze badges

2 Comments

MatthewMartin Over a year ago

Yes, that would work for Console apps, but event for Console apps you might want to aggregate them first in a collection, or else you get interleaved results to the Console window.

MichaC Over a year ago

The Console.WriteLine commands are running synchronously on the main thread and it will print the results in the order they were defined in the original List after Parallel.ForEach finishes processing all the list items and returns. If I were calling WriteLine from within the Parallel.ForEach then yes the results would be interleaved.

Collectives™ on Stack Overflow

How do I collect return values from Parallel.ForEach?

6 Answers 6

10 Comments

6 Comments

3 Comments

2 Comments

2 Comments

2 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

6 Answers 6

10 Comments

6 Comments

3 Comments

2 Comments

2 Comments

2 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related