12

我需要在工作流中引入重试策略。假设有 3 个块以这种方式连接：

var executionOptions = new ExecutionDataflowBlockOptions { MaxDegreeOfParallelism = 3 };
var buffer = new BufferBlock<int>();
var processing = new TransformBlock<int, int>(..., executionOptions);
var send = new ActionBlock<int>(...);

buffer.LinkTo(processing);
processing.LinkTo(send);

所以有一个缓冲区来累积数据，然后将其发送到一次处理不超过 3 个项目的转换块，然后将结果发送到操作块。

在处理转换块期间可能会出现瞬态错误，如果错误多次出现瞬态，我想重试该块。

我知道块通常是不可重试的（传递到块中的委托可以重试）。一种选择是包装传递的委托以支持重试。

我也知道有一个非常好的库TransientFaultHandling.Core可以为瞬态故障提供重试机制。这是一个很棒的图书馆，但对我来说不是。如果我将传递给转换块的委托包装到RetryPolicy.ExecuteAsync方法中，则转换块内的消息将被锁定，并且在重试完成或失败之前，转换块将无法接收新消息。想象一下，如果所有 3 条消息都进入重试（假设下一次重试将在 2 分钟内）并且失败，则转换块将被卡住，直到至少有一条消息离开转换块。

我看到的唯一解决方案是扩展TranformBlock（实际上ITargetBlock也足够了），并手动重试（比如从这里开始）：

do
 {
    try { return await transform(input); }
    catch
    { 
        if( numRetries <= 0 ) throw;
        else Task.Delay(timeout).ContinueWith(t => processing.Post(message));
    }
 } while( numRetries-- > 0 );

ig 将消息再次延迟放入转换块中，但在这种情况下，重试上下文（剩余的重试次数等）也应传递到此块中。听起来太复杂了...

有没有人看到一种更简单的方法来为工作流块实施重试策略？

4

3 回答 3

16

我认为您几乎必须这样做，您必须跟踪消息的剩余重试次数，并且您必须以某种方式安排重试尝试。

但是您可以通过将其封装在单独的方法中来使其变得更好。就像是：

// it's a private class, so public fields are okay
private class RetryingMessage<T>
{
    public T Data;
    public int RetriesRemaining;
    public readonly List<Exception> Exceptions = new List<Exception>();
}

public static IPropagatorBlock<TInput, TOutput>
    CreateRetryingBlock<TInput, TOutput>(
    Func<TInput, Task<TOutput>> transform, int numberOfRetries,
    TimeSpan retryDelay, Action<IEnumerable<Exception>> failureHandler)
{
    var source = new TransformBlock<TInput, RetryingMessage<TInput>>(
        input => new RetryingMessage<TInput>
        { Data = input, RetriesRemaining = numberOfRetries });

    // TransformManyBlock, so that we can propagate zero results on failure
    TransformManyBlock<RetryingMessage<TInput>, TOutput> target = null;
    target = new TransformManyBlock<RetryingMessage<TInput>, TOutput>(
        async message =>
        {
            try
            {
                return new[] { await transform(message.Data) };
            }
            catch (Exception ex)
            {
                message.Exceptions.Add(ex);
                if (message.RetriesRemaining == 0)
                {
                    failureHandler(message.Exceptions);
                }
                else
                {
                    message.RetriesRemaining--;
                    Task.Delay(retryDelay)
                        .ContinueWith(_ => target.Post(message));
                }
                return null;
            }
        });

    source.LinkTo(
        target, new DataflowLinkOptions { PropagateCompletion = true });

    return DataflowBlock.Encapsulate(source, target);
}

我添加了代码来跟踪异常，因为我认为不应该忽略失败，它们至少应该被记录下来。

此外，此代码在完成时不能很好地工作：如果有重试等待他们的延迟并且您Complete()阻塞，它将立即完成并且重试将丢失。如果这对您来说是个问题，您将必须跟踪未完成的重试并在完成target时source完成并且没有重试正在等待。

于 2013-07-04T13:04:24.583 回答
3

除了 svick 的出色回答之外，还有其他几个选项：

  1. 您可以使用TransientFaultHandling.Core- 只需设置MaxDegreeOfParallelism为Unbounded以便其他消息可以通过。
  2. 您可以修改块输出类型以包括失败指示和重试计数，并创建一个数据流循环，将一个过滤器传递给LinkTo它以检查是否需要再次重试。这种方法更复杂；如果它正在重试，您必须为您的块添加延迟，并添加一个TransformBlock以删除网格其余部分的失败/重试信息。
于 2013-07-05T02:58:17.397 回答
2

以下是在这些假设下运行CreateRetryTransformBlock的两种方法：CreateRetryActionBlock

  1. 调用者希望处理所有项目，即使其中一些项目反复失败。
  2. 调用者有兴趣了解所有发生的异常，即使是最终成功的项目（不适用于CreateRetryActionBlock）。
  3. 调用者可能希望设置总重试次数的上限，在此之后块应转换为故障状态。
  4. 调用者希望能够在与重试功能相关的选项之上 设置普通块的所有可用选项，包括MaxDegreeOfParallelism、和BoundedCapacity， 。CancellationTokenEnsureOrdered

下面的实现使用 aSemaphoreSlim来控制第一次尝试的操作与在延迟持续时间过去后重试的先前错误操作之间的并发级别。

public class RetryExecutionDataflowBlockOptions : ExecutionDataflowBlockOptions
{
    /// <summary>The limit after which an item is returned as failed.</summary>
    public int MaxAttemptsPerItem { get; set; } = 1;
    /// <summary>The delay duration before retrying an item.</summary>
    public TimeSpan RetryDelay { get; set; } = TimeSpan.Zero;
    /// <summary>The limit after which the block transitions to a faulted
    /// state (unlimited is the default).</summary>
    public int MaxRetriesTotal { get; set; } = -1;
}

public readonly struct RetryResult<TInput, TOutput>
{
    public readonly TInput Input { get; }
    public readonly TOutput Output { get; }
    public readonly bool Success { get; }
    public readonly Exception[] Exceptions { get; }

    public bool Failed => !Success;
    public Exception FirstException => Exceptions != null ? Exceptions[0] : null;
    public int Attempts =>
        Exceptions != null ? Exceptions.Length + (Success ? 1 : 0) : 1;

    public RetryResult(TInput input, TOutput output, bool success,
        Exception[] exceptions)
    {
        Input = input;
        Output = output;
        Success = success;
        Exceptions = exceptions;
    }
}

public class RetryLimitException : Exception
{
    public RetryLimitException(string message, Exception innerException)
        : base(message, innerException) { }
}

public static IPropagatorBlock<TInput, RetryResult<TInput, TOutput>>
    CreateRetryTransformBlock<TInput, TOutput>(
    Func<TInput, Task<TOutput>> transform,
    RetryExecutionDataflowBlockOptions dataflowBlockOptions)
{
    if (transform == null) throw new ArgumentNullException(nameof(transform));
    if (dataflowBlockOptions == null)
        throw new ArgumentNullException(nameof(dataflowBlockOptions));
    int maxAttemptsPerItem = dataflowBlockOptions.MaxAttemptsPerItem;
    int maxRetriesTotal = dataflowBlockOptions.MaxRetriesTotal;
    TimeSpan retryDelay = dataflowBlockOptions.RetryDelay;
    if (maxAttemptsPerItem < 1) throw new ArgumentOutOfRangeException(
        nameof(dataflowBlockOptions.MaxAttemptsPerItem));
    if (maxRetriesTotal < -1) throw new ArgumentOutOfRangeException(
        nameof(dataflowBlockOptions.MaxRetriesTotal));
    if (retryDelay < TimeSpan.Zero) throw new ArgumentOutOfRangeException(
        nameof(dataflowBlockOptions.RetryDelay));
    var cancellationToken = dataflowBlockOptions.CancellationToken;

    var exceptionsCount = 0;
    var semaphore = new SemaphoreSlim(
        dataflowBlockOptions.MaxDegreeOfParallelism);

    async Task<(TOutput, Exception)> ProcessOnceAsync(TInput item)
    {
        await semaphore.WaitAsync(); // Preserve the SynchronizationContext
        try
        {
            var result = await transform(item).ConfigureAwait(false);
            return (result, null);
        }
        catch (Exception ex)
        {
            if (maxRetriesTotal != -1)
            {
                if (Interlocked.Increment(ref exceptionsCount) > maxRetriesTotal)
                {
                    throw new RetryLimitException($"The max retry limit " +
                        $"({maxRetriesTotal}) has been reached.", ex);
                }
            }
            return (default, ex);
        }
        finally
        {
            semaphore.Release();
        }
    }

    async Task<Task<RetryResult<TInput, TOutput>>> ProcessWithRetryAsync(
        TInput item)
    {
        // Creates a two-stages operation. Preserves the context on every await.
        var (result, firstException) = await ProcessOnceAsync(item);
        if (firstException == null) return Task.FromResult(
            new RetryResult<TInput, TOutput>(item, result, true, null));
        return RetryStageAsync();

        async Task<RetryResult<TInput, TOutput>> RetryStageAsync()
        {
            var exceptions = new List<Exception>();
            exceptions.Add(firstException);
            for (int i = 2; i <= maxAttemptsPerItem; i++)
            {
                await Task.Delay(retryDelay, cancellationToken);
                var (result, exception) = await ProcessOnceAsync(item);
                if (exception != null)
                    exceptions.Add(exception);
                else
                    return new RetryResult<TInput, TOutput>(item, result,
                        true, exceptions.ToArray());
            }
            return new RetryResult<TInput, TOutput>(item, default, false,
                exceptions.ToArray());
        };
    }

    // The input block awaits the first stage of each operation
    var input = new TransformBlock<TInput, Task<RetryResult<TInput, TOutput>>>(
        item => ProcessWithRetryAsync(item), dataflowBlockOptions);

    // The output block awaits the second (and final) stage of each operation
    var output = new TransformBlock<Task<RetryResult<TInput, TOutput>>,
        RetryResult<TInput, TOutput>>(t => t, dataflowBlockOptions);

    input.LinkTo(output, new DataflowLinkOptions { PropagateCompletion = true });

    // In case of failure ensure that the input block is faulted too,
    // so that its input/output queues are emptied, and any pending
    // SendAsync operations are aborted
    PropagateFailure(output, input);

    return DataflowBlock.Encapsulate(input, output);

    async void PropagateFailure(IDataflowBlock block1, IDataflowBlock block2)
    {
        try { await block1.Completion.ConfigureAwait(false); }
        catch (Exception ex) { block2.Fault(ex); }
    }
}

public static ITargetBlock<TInput> CreateRetryActionBlock<TInput>(
    Func<TInput, Task> action,
    RetryExecutionDataflowBlockOptions dataflowBlockOptions)
{
    if (action == null) throw new ArgumentNullException(nameof(action));
    var block = CreateRetryTransformBlock<TInput, object>(async input =>
    {
        await action(input).ConfigureAwait(false); return null;
    }, dataflowBlockOptions);
    var nullTarget = DataflowBlock.NullTarget<RetryResult<TInput, object>>();
    block.LinkTo(nullTarget);
    return block;
}
于 2020-05-24T13:05:18.007 回答