Azure Durable Functions(dotnet-isolated)：如何实现基于异常类型的重试逻辑？

一、从AggregateException中提取真实异常类型 #

retryContext.LastFailure返回的AggregateException是对底层真实异常的包装，你可以通过以下方式获取目标TaskFailedException：

• 使用AggregateException.Flatten()展开嵌套异常，再遍历查找目标类型
• 直接遍历InnerExceptions集合筛选

示例代码：

private bool IsTransientException(Exception exception)
{
    if (exception is AggregateException aggEx)
    {
        // 展开所有嵌套异常，判断是否存在瞬态TaskFailedException
        return aggEx.Flatten().InnerExceptions.Any(ex => 
            ex is TaskFailedException taskFailedEx && taskFailedEx.IsTransient());
    }
    // 处理非包装的直接异常
    return exception is TaskFailedException taskFailedEx && taskFailedEx.IsTransient();
}

二、实现同时支持参数配置与异常判断的自定义重试策略 #

针对现有两种方式的局限性，你可以手动实现IRetryPolicy接口，同时控制重试参数和异常判断逻辑，完全满足需求。

1. 自定义RetryPolicy实现 #

internal class CustomTransientRetryPolicy : IRetryPolicy
{
    private readonly int _maxAttempts;
    private readonly TimeSpan _firstRetryInterval;
    private readonly TimeSpan _maxRetryInterval;
    private readonly double _backoffCoefficient;

    // 通过构造函数传入重试配置参数
    public CustomTransientRetryPolicy(int maxAttempts, TimeSpan firstRetryInterval, TimeSpan maxRetryInterval, double backoffCoefficient)
    {
        _maxAttempts = maxAttempts;
        _firstRetryInterval = firstRetryInterval;
        _maxRetryInterval = maxRetryInterval;
        _backoffCoefficient = backoffCoefficient;
    }

    // 实现指数退避的重试间隔计算逻辑
    public Task<TimeSpan?> GetNextRetryIntervalAsync(RetryContext retryContext)
    {
        // 超过最大尝试次数则终止重试
        if (retryContext.AttemptNumber >= _maxAttempts)
        {
            return Task.FromResult<TimeSpan?>(null);
        }

        // 计算退避间隔，不超过最大重试间隔上限
        var retryInterval = _firstRetryInterval * Math.Pow(_backoffCoefficient, retryContext.AttemptNumber - 1);
        var cappedInterval = TimeSpan.FromMilliseconds(Math.Min(retryInterval.TotalMilliseconds, _maxRetryInterval.TotalMilliseconds));
        
        return Task.FromResult<TimeSpan?>(cappedInterval);
    }

    // 实现异常判断逻辑，仅重试瞬态异常
    public Task<bool> ShouldRetryAsync(RetryContext retryContext)
    {
        if (retryContext.LastFailure == null)
        {
            return Task.FromResult(false);
        }

        var shouldRetry = IsTransientException(retryContext.LastFailure);
        return Task.FromResult(shouldRetry);
    }
}

2. 在Durable Task中应用自定义策略 #

配置TaskOptions时，直接通过FromRetryPolicy传入自定义策略实例：

// 初始化自定义重试策略，传入配置参数
var retryPolicy = new CustomTransientRetryPolicy(
    maxAttempts: 5,
    firstRetryInterval: TimeSpan.FromSeconds(10),
    maxRetryInterval: TimeSpan.FromSeconds(20),
    backoffCoefficient: 2);

// 在调用活动函数/子编排时应用该策略
var taskOptions = TaskOptions.FromRetryPolicy(retryPolicy);
await context.CallActivityAsync<string>("YourTargetActivity", input, taskOptions);

说明 #

该方案适配Microsoft.Azure.Functions.Worker.Extensions.DurableTask 1.1.2版本，避开了过时的HandleAsync方法，同时实现了：

• 灵活配置重试次数、间隔、退避系数等参数
• 精准识别异常类型，仅对瞬态异常执行重试

内容的提问来源于stack exchange，提问作者Alex