Parallel.For fail (C#)

Question

I wrote some code:

class Program
    {
        public const int count = 3000;
        static List<int> list = new List<int>();
        static void DoWork(int i)
        {            
            list.Add(i);
        }        
        static void Main(string[] args)
        {
            while (true)
            {

                Stopwatch s = new Stopwatch();
                s.Start();
                Parallel.For(0, count + 1, DoWork);            
                s.Stop();
                Console.WriteLine("\n Elapsed: " + s.Elapsed.ToString());
                Console.WriteLine("Expected: {0}", count + 1);
                Console.WriteLine("count: {0}", list.Count);
                Console.ReadKey();
                list = new List<int>(); 
            }
        }
    }

but results are not expected(

Not all of the cycles are finished before Console.WriteLine calls

What is the problem with using Parallel.For?

Answer 1

You're running into what's known as a Race Condition. Since the List collection in .Net isn't thread safe, it's operations such as Add() aren't atomic. Basically a call to Add() on one thread can destroy another thread's Add() before it is complete. You need a thread-safe concurrent collection for your code.

Try this:

using System.Threading.Tasks;
class Program
{

    public const int count = 3000;
    static ConcurrentBag<int> bag = new ConcurrentBag<int>();
    static void DoWork(int i)
    {
        bag.Add(i);
    }
    static void Main(string[] args)
    {
        while (true)
        {

            Stopwatch s = new Stopwatch();
            s.Start();
            Parallel.For(0, count + 1, DoWork);
            s.Stop();
            Console.WriteLine("\n Elapsed: " + s.Elapsed.ToString());
            Console.WriteLine("Expected: {0}", count + 1);
            Console.WriteLine("count: {0}", bag.Count);
            Console.ReadKey();
            bag = new ConcurrentBag<int>();
        }
    }
}

The ConcurrentBag is the closest thing to a thread-safe list. Just remember since we are dealing with unknown scheduling, the integers won't be in order.

Answer 2

The List<> class is not thread save. You can't modify it in a parallel loop (Without problems). Use a collection form the System.Collections.Concurrent namespace