I have a question about async programming and Task.WhenAll(). I have a code snippet that downloads a folder from google drive and it works as expected when i am debugging my code. However when I run the application without debugger the download function takes atleast 4x the time it takes when it runs with the debugger. I also get crash logs with TaskCancelled exceptions (System.Threading.Tasks.TaskCanceledException: A task was canceled.) which do not happen with the debugger attached. What needs to be changed for the code to work as expected without debugger attached. NB this snippet downloads - 1000 files in about 22-25 seconds with debugger and 2min without debugger.
public static async Task<bool> DownloadFolder(CloudDataModel.File file, string path, params string[] exclude)
{
try
{
if (file != null && !string.IsNullOrEmpty(file.id))
{
List<string> toExclude = new List<string>();
if(exclude != null)
{
toExclude = exclude.ToList();
}
List<Task> downloadFilesTask = new List<Task>();
var files = await file.GetFiles();
foreach (var f in files)
{
var task = f.Download(path);
downloadFilesTask.Add(task);
}
var folders = await file.GetFoldersAsync();
foreach (var folder in folders)
{
if (toExclude.Contains(folder.name))
{
continue;
}
Task task = null;
if (path.Equals(Statics.ProjectFolderName))
{
task = DownloadFolder(folder, folder.name);
}
else
{
task = DownloadFolder(folder, Path.Combine(path, folder.name));
}
downloadFilesTask.Add(task);
}
var array = downloadFilesTask.ToArray();
await Task.WhenAll(array);
return true;
}
}
catch (Exception e)
{
Crashes.TrackError(e);
}
return false;
}
Edit
after some more trial and error the fault has been identified. the downloading of the file was the cause of the unexpected behaviour
public static async Task<StorageFile> DownloadFile(CloudDataModel.File file, string destinationFolder)
{
try
{
if (file != null)
{
Debug.WriteLine($"start download {file.name}");
if (file.mimeType == Statics.GoogleDriveFolderMimeType)
{
Debug.WriteLine($"did not download resource, resource was folder instead of file. mimeType: {file.mimeType}");
return null;
}
var endpoint = Path.Combine(DownloadFileEndpoint, $"{file.id}?alt=media");
// this would cause the unexpected behaviour
HttpResponseMessage response = await client.GetAsync(endpoint);
StorageFile downloadedFile;
using (Stream streamToReadFrom = await response.Content.ReadAsStreamAsync())
{
var memstream = new MemoryStream();
StreamReader reader = new StreamReader(streamToReadFrom);
streamToReadFrom.Position = 0;
await streamToReadFrom.CopyToAsync(memstream);
downloadedFile = await fileHandler.SaveDownloadedCloudFile(memstream, file, destinationFolder);
Debug.WriteLine($"download finished {file.name}");
}
return downloadedFile;
}
return null;
}
catch (Exception e)
{
Crashes.TrackError(e);
return null;
}
}
After setting a timeout to the client (System.Net.Http.HttpClient) the code executed as expected.
client.Timeout = new TimeSpan(0,0,5);
CodePudding user response:
So after some trial and error the downloading of the file was identified as the cause of the problem.
var task = f.Download(path);
the Download function implements a System.Net.Http.HttpClient wich was initialized without any kind of timeout. when one of the tasks had a timeout, all other tasks would also not execute as expected. to fix this issue a timeout of 5 seconds was added to the HttpClient object.
private static void InitializeClient()
{
HttpClientHandler handler = new HttpClientHandler { AllowAutoRedirect = true };
client = new HttpClient(handler);
client.DefaultRequestHeaders.Authorization = new System.Net.Http.Headers.AuthenticationHeaderValue("Bearer", AccountManager.LoggedInAccount.Token);
client.Timeout = new TimeSpan(0,0,5);
}
After this change the code would execute as expected.
CodePudding user response:
My guess is that without the debugger the parallelism increases, and so the remote server is overflowed by requests and can't perform optimally. Your code does not have any provision for limiting the degree of concurrency/parallelism. One possible way to solve this problem is to use a SemaphoreSlim
, as shown in this question (among many others).