How to run a lot of Tasks in batches of smaller size?

Question

I have a lot of operations what i want to run async. I do:

var tasks = new List<Task<bool>>();

for(var i=0;i<1000;i++){
  tasks.Add(CreateGeoreferencedImageAsync(properties, scaleIndex, currentXmin, currentYmin, currentXmax, currentYmax));
}

while (tasks.Count > 0)
{
   var bunch = tasks.Take(4).ToList();
   bool[] firstFinishedTask = await Task.WhenAll(bunch);
   tasks.RemoveRange(0,4);
}

But i see that WhenAll execute all Task from tasks not only from bunch.
What i missed?

UPDATE

private Task<bool> CreateGeoreferencedImageAsync(ImageGenerationProperties 
properties, int scaleIndex,
        double currentXmin, double currentYmin, double currentXmax, double currentYmax)
    {
        return Task.Run(() =>
            {
                return CreateGeoreferencedImage(properties, scaleIndex, currentXmin, currentYmin, currentXmax,
                    currentYmax);
            });
    }

Your list contains tasks that are already started. There is absolutely no reason to wait for a couple of them at a time. — FCin, Oct 01 '18 at 11:48
So maybe the correct answer was that he should not create the Tasks with `Task.Run` but `new Task`. Maybe this is helpful: https://stackoverflow.com/questions/14075029/have-a-set-of-tasks-with-only-x-running-at-a-time — Tim Schmelter, Oct 01 '18 at 11:53
Why don't you simply `Task.WaitAll(tasks)`? Whar are you trying to avoid? — Gabriel, Oct 01 '18 at 11:54
I added method code. And i want to run just a part of Task couse i have not enough memory to run them all. — Kliver Max, Oct 01 '18 at 11:57
Possible duplicate of [How to limit the amount of concurrent async I/O operations?](https://stackoverflow.com/questions/10806951/how-to-limit-the-amount-of-concurrent-async-i-o-operations) — Joe Phillips, Oct 01 '18 at 11:58
@JoePhillips - no that is _not_ a (good) duplicate. This question is not about I/O, and that is crucial. — bommelding, Oct 01 '18 at 12:01
Just use Microsoft's Dataflow api: https://learn.microsoft.com/en-us/dotnet/standard/parallel-programming/dataflow-task-parallel-library — zaitsman, Oct 01 '18 at 12:02
@KliverMax - you are not very clear about what you want with the returned booleans. Just disregard them? — bommelding, Oct 01 '18 at 12:04
@bommelding I got errors when used `void` and do not remove bool return value. — Kliver Max, Oct 01 '18 at 12:07
So, there is no return? Are the GeoreferencedImages retrieved from one place and stored in another? Could any of that I/O be made async? — bommelding, Oct 01 '18 at 12:09
I think that i understand idea that i gonna create new Tasks just after previous banch finish to work. — Kliver Max, Oct 01 '18 at 12:18

score 3 · Accepted Answer · answered Oct 01 '18 at 11:48

3

WhenAll doesn't execute the tasks. It waits for them to finish. The execution usually starts as soon as the task is created, in your case inside CreateGeoreferencedImageAsync.

answered Oct 01 '18 at 11:48

Daniel Hilgarth

171,043
40
335
443

It is mean that i gonna create new banch of Tasks after previous banch finish to work? – Kliver Max Oct 01 '18 at 11:59

score 3 · Answer 2 · edited Oct 01 '18 at 12:17

3

Forget about Tasks and especially about Task.Run().

This problem is best solved with Parallel.ForEach(). Or Linq's AsParallel().

Parallel.For(0, 1000, 
  new ParallelOptions { MaxDegreeOfParallelism = 4 },
  i => CreateGeoreferencedImage(properties, scaleIndex, 
      currentXmin, currentYmin, currentXmax,  currentYmax));

edited Oct 01 '18 at 12:17

Joe Phillips

49,743
32
103
159

answered Oct 01 '18 at 11:57

bommelding

2,969
9
14

I believe this is an older way to do it. Semaphores would be the newer way. I could be wrong, too – Joe Phillips Oct 01 '18 at 11:59
1

Semaphores are way older than the Parallel class. About half a century. – bommelding Oct 01 '18 at 12:00
SemaphoreSlim isn't though. http://blog.briandrupieski.com/throttling-asynchronous-methods-in-csharp – Joe Phillips Oct 01 '18 at 12:02
Now is your chance to explain why to use TPL instead of the other options then. I think I understand why but I wouldn't expect others to – Joe Phillips Oct 01 '18 at 12:04
No, the question isn't finished yet. – bommelding Oct 01 '18 at 12:05
I agree this is the answer but I feel its important to explain why. My understanding is that it is a cpu bound (non async) "task" that is occurring. If it were actually async then semaphoreslim would make more sense here – Joe Phillips Oct 01 '18 at 12:19
@JoePhillips - it's probably not really CPU bound but as long as the interface of the core method is not awaitable, Tasks and WhenAll are a clumsy workaround. And I still don't see the use for semaphores here. With or without Tasks. – bommelding Oct 01 '18 at 12:33
@bommelding The main reason to use SemaphoreSlim here would be that Parallel.For will block threads. SemaphoreSlim can be used asynchronously. Maybe you can do the same with Parallel.For as well? – Joe Phillips Oct 01 '18 at 15:48
@JoePhillips - Yes, the Parallel.For() is blocking as a whole. When that is a problem, wrap it in a single Task.Run(). But there is nothing in the question that indicates that this is desired. – bommelding Oct 03 '18 at 06:26

How to run a lot of Tasks in batches of smaller size?

2 Answers2