问题
I have a class that is responsible for retrieving a product availability by making call to a legacy class. This legacy class itself internally collects product data by making BLOCKING network calls. Note that I cannot modify code of legacy API. Since all products are independent to each other, I would like to parallelise collecting the information without creating any unnecessary threads and also not blocking thread that gets blocked on calling this legacy API. With this background here are my basic classes.
class Product
{
public int ID { get; set; }
public int VendorID { get; set; }
public string Name { get; set; }
}
class ProductSearchResult
{
public int ID { get; set; }
public int AvailableQuantity { get; set; }
public DateTime ShipDate { get; set; }
public bool Success { get; set; }
public string Error { get; set; }
}
class ProductProcessor
{
List<Product> products;
private static readonly SemaphoreSlim mutex = new SemaphoreSlim(2);
CancellationTokenSource cts = new CancellationTokenSource();
public ProductProcessor()
{
products = new List<Product>()
{
new Product() { ID = 1, VendorID = 100, Name = "PC" },
new Product() { ID = 2, VendorID = 101, Name = "Tablet" },
new Product() { ID = 3, VendorID = 100, Name = "Laptop" },
new Product() { ID = 4, VendorID = 102, Name = "GPS" },
new Product() { ID = 5, VendorID = 107, Name = "Mars Rover" }
};
}
public async void Start()
{
Task<ProductSearchResult>[] tasks = new Task<ProductSearchResult>[products.Count];
Parallel.For(0, products.Count(), async i =>
{
tasks[i] = RetrieveProductAvailablity(products[i].ID, cts.Token);
});
Task<ProductSearchResult> results = await Task.WhenAny(tasks);
// Logic for waiting on indiviaul tasks and reporting results
}
private async Task<ProductSearchResult> RetrieveProductAvailablity(int productId, CancellationToken cancellationToken)
{
ProductSearchResult result = new ProductSearchResult();
result.ID = productId;
if (cancellationToken.IsCancellationRequested)
{
result.Success = false;
result.Error = "Cancelled.";
return result;
}
try
{
await mutex.WaitAsync();
if (cancellationToken.IsCancellationRequested)
{
result.Success = false;
result.Error = "Cancelled.";
return result;
}
LegacyApp app = new LegacyApp();
bool success = await Task.Run(() => app.RetrieveProductAvailability(productId));
if (success)
{
result.Success = success;
result.AvailableQuantity = app.AvailableQuantity;
result.ShipDate = app.ShipDate;
}
else
{
result.Success = false;
result.Error = app.Error;
}
}
finally
{
mutex.Release();
}
return result;
}
}
Given that I am trying to wrap async over a synchronous API, I have two questions.
- With use of Parallel.For and wrapping Legay API call within a Task.Run, am I creating any unnecessary threads that could have been avoided without blocking calling thread as we will use this code in UI.
- Is this code still look thread safe.
回答1:
The compiler will give you warnings about your async lambda. Read it carefully; it's telling you that it's not asynchronous. There's no point in using async there. Also, do not use async void.
Since your underlying API is blocking - and there's no way to change that - asynchronous code isn't an option. I'd recommend either using several Task.Run calls or Parallel.For, but not both. So let's use parallel. Actually, let's use Parallel LINQ since you're transforming a sequence.
There's no point in making RetrieveProductAvailablity asynchronous; it's only doing blocking work except for the throttling, and the parallel approach has more natural throttling support. This leaves your method looking like:
private ProductSearchResult RetrieveProductAvailablity(int productId, CancellationToken cancellationToken)
{
... // no mutex code
LegacyApp app = new LegacyApp();
bool success = app.RetrieveProductAvailability(productId);
... // no mutex code
}
You can then do parallel processing as such:
public void Start()
{
ProductSearchResult[] results = products.AsParallel().AsOrdered()
.WithCancellation(cts.Token).WithDegreeOfParallelism(2)
.Select(product => RetrieveProductAvailability(product.ID, cts.Token))
.ToArray();
// Logic for waiting on indiviaul tasks and reporting results
}
From your UI thread, you can call the method using Task.Run:
async void MyUiEventHandler(...)
{
await Task.Run(() => processor.Start());
}
This keeps your business logic clean (only synchronous/parallel code), and the responsibility for moving this work off the UI thread (using Task.Run) belongs to the UI layer.
Update: I added a call to AsOrdered to ensure the results array has the same order as the products sequence. This may or may not be necessary, but since the original code preserved order, this code does now too.
Update: Since you need to update the UI after every retrieval, you should probably use Task.Run for each one instead of AsParallel:
public async Task Start()
{
var tasks = products.Select(product =>
ProcessAvailabilityAsync(product.ID, cts.Token));
await Task.WhenAll(tasks);
}
private SemaphoreSlim mutex = new SempahoreSlim(2);
private async Task ProcessAvailabilityAsync(int id, CancellationToken token)
{
await mutex.WaitAsync();
try
{
var result = await RetrieveProductAvailability(id, token);
// Logic for reporting results
}
finally
{
mutex.Release();
}
}
回答2:
am I creating any unnecessary threads that could have been avoided without blocking calling thread as we will use this code in UI.
Yes. Your code spins new threads via Parallel.ForEach, and then again internally inside RetrieveProductAvailablity. There is no need for that.
async-await and Parallel.ForEach don't really play nice together, as it converts your async lambda into an async void method instead of async Task.
What i would recommend is to drop the Parallel.ForEach and the wrapped sync call and do that following:
Change your method call from async to sync (as it really isn't async at all):
private ProductSearchResult RetrieveProductAvailablity(int productId,
CancellationToken
cancellationToken)
Instead of this:
bool success = await Task.Run(() => app.RetrieveProductAvailability(productId));
Invoke the method call synchronously:
bool success = app.RetrieveProductAvailability(productId));
And then explicitly invoke Task.Run on all of them:
var productTasks = products.Select(product => Task.Run(() =>
RetrieveProductAvailablity(product.ID, cts.Token))
await Task.WhenAll(productTasks);
Generally, it's not recommended to expose async wrappers over sync methods
来源：https://stackoverflow.com/questions/28071444/wrapping-blocking-calls-to-be-async-for-better-thread-reuse-and-responsive-ui