C# 中带有重试逻辑的 Parallel.ForEach 循环

Question

I am using Parallel.ForEach to download multiple files in C# from google bucket to folder location. I'm using retry logic so it can retry downloading files in case files download fails during downloading. How can I apply retry logic for each file or each thread in Parallel.ForEach loop.

Parallel.ForEach(listFiles, objectName =>            
{
    retryCount = 0;                        
    countOfFiles++;
    downloadSuccess = false;
    bucketFileName = Path.GetFileName(objectName.Name);
    guidFolderPath = tempFolderLocation + "\\" + bucketFileName;

    while (retryCount < retryCountInput && downloadSuccess == false)
    {
        try
        {
            FileStream fs = new FileStream(guidFolderPath, FileMode.Create, FileAccess.Write, FileShare.Write);
            using (fs)
            {                                               
                storage.DownloadObjectAsync(bucketName, objectName.Name, fs, option, cancellationToken, progress).Wait();
            }
        }
        catch (Exception ex)
        {
            Console.WriteLine("Exception occured while downloading file: " + ex.ToString());                   
            Thread.Sleep(RetryInterval(retryCount, minBackoffTimeSpan, maxBackoffTimeSpan, deltaBackoffTimeSpan));
            retryCount++;

        }
    }
}

Answer 1

I would change it to tasks and use the async. This way your Thread.Sleep doesn't block a threadpool thread. The Parallel.ForEach is for CPU bound work.

Something like: (I'm unable to compile/test this without the rest of your code)

int retryCountInput = 5;
var tasks = new List<Task>();

foreach (var file in listFiles)
{
    var task = Task.Run(async () =>
    {
        // make it local
        int retryCount = 0;
        string bucketFileName = Path.GetFileName(objectName.Name);
        string guidFolderPath = tempFolderLocation + "\\" + bucketFileName;

        while (retryCount < retryCountInput)
        {
            try
            {
                using (var fs = new FileStream(guidFolderPath, FileMode.Create, FileAccess.Write, FileShare.Write))
                    // Use await here, instead of `Wait()` so this threadpool thread
                    // can be used for other tasks.
                    await storage.DownloadObjectAsync(bucketName, objectName.Name, fs, option, cancellationToken, progress);

                break;
            }
            catch (Exception ex)
            {
                Console.WriteLine("Exception occured while downloading file: " + ex.ToString());

                // Use Task.Delay here, so this thread is 'released'
                await Task.Delay(RetryInterval(retryCount, minBackoffTimeSpan, maxBackoffTimeSpan, deltaBackoffTimeSpan));
                retryCount++;
            }
        }
    });
    tasks.Add(task);
}
await Task.WhenAll(tasks);

Answer 2

I have modified my code and removed Parallel.ForEach instead using foreach loop to iterate through files. But now, I'm not able to find all files in downloaded path, though logs shows all files got downloaded. Number of downloaded files in download path changes and this behavior seems random. Can I use Task.Run for I/O operations?

var tasks = new List<Task>();
foreach (var objectName in listFiles)
{
    var task = Task.Run(() =>
    {
        downloadSuccess = false;
        bucketFileName = Path.GetFileName(objectName.Name);
        guidFolderPath = tempFolderLocation + "\\" + bucketFileName;

        var maxRetryAttempts = 3;
        var pauseBetweenFailures = TimeSpan.FromSeconds(2);
        RetryHelper.RetryOnException(maxRetryAttempts, pauseBetweenFailures, async () =>
        {
            FileStream fs = new FileStream(guidFolderPath, FileMode.Create,
                FileAccess.Write, FileShare.Write);
            using (fs)
            {
                var progress = new Progress<IDownloadProgress>(
                    p =>
                    {
                        DownloadProgress(p, retryCount, objectName.Name);
                    });

                await client.DownloadObjectAsync(bucketName, objectName.Name,
                    fs, option, cancellationToken.Token, progress);
            }
        });
    });
    tasks.Add(task);
}
await Task.WhenAll(tasks);