Question

我有以下政策：

var retryPolicy = Policy.Handle<Exception>(e => (e is HttpRequestException || e.InnerException is HttpRequestException)).WaitAndRetry(
                retryCount: maxRetryCount,
                sleepDurationProvider: attempt => TimeSpan.FromSeconds(Math.Pow(2, attempt)),
                onRetry: (exception, calculatedWaitDuration, retryCount, context) =>
                {
                    Log.Error($"Retry => Count: {retryCount}, Wait duration: {calculatedWaitDuration}, Policy Wrap: {context.PolicyWrapKey}, Policy: {context.PolicyKey}, Endpoint: {context.OperationKey}, Exception: {exception}.");
                });

var circuitBreaker = Policy.Handle<Exception>(e => (e is HttpRequestException || e.InnerException is HttpRequestException)).CircuitBreaker(maxExceptionsBeforeBreaking, TimeSpan.FromSeconds(circuitBreakDurationSeconds), onBreak, onReset);

var sharedBulkhead = Policy.Bulkhead(maxParallelizations, maxQueuingActions, onBulkheadRejected);

var fallbackForCircuitBreaker = Policy<bool>
             .Handle<BrokenCircuitException>()
             .Fallback(
                 fallbackValue: false,
                 onFallback: (b, context) =>
                 {
                     Log.Error($"Operation attempted on broken circuit => Policy Wrap: {context.PolicyWrapKey}, Policy: {context.PolicyKey}, Endpoint: {context.OperationKey}");
                 }
             );

            var fallbackForAnyException = Policy<bool>
                .Handle<Exception>()
                .Fallback(
                    fallbackAction: (context) => { return false; },
                    onFallback: (e, context) =>
                    {
                        Log.Error($"An unexpected error occured => Policy Wrap: {context.PolicyWrapKey}, Policy: {context.PolicyKey}, Endpoint: {context.OperationKey}");
                    }
                );

var resilienceStrategy = Policy.Wrap(retryPolicy, circuitBreaker, sharedBulkhead);
            var policyWrap = fallbackForAnyException.Wrap(fallbackForCircuitBreaker.Wrap(resilienceStrategy));

public bool CallApi(ChangeMapModel changeMessage)
    {
        var httpClient = new HttpClient();
        var endPoint = changeMessage.EndPoint;
        var headers = endPoint.Headers;
        if (headers != null)
        {
            foreach (var header in headers)
            {
                if (header.Contains(':'))
                {
                    var splitHeader = header.Split(':');
                    httpClient.DefaultRequestHeaders.Add(splitHeader[0], splitHeader[1]); 
                }
            } 
        }

        var res = httpClient.PostAsync(endPoint.Uri, null);
        var response = res.Result;
        response.EnsureSuccessStatusCode();
        return true;
    }

我这样执行政策：

policyWrap.Execute((context) => CallApi(changeMessage), new Context(endPoint));

问题是，在开路执行操作时，CircuitBreaker回调没有被击中。

我希望通过策略放置API调用，并以HttpRequestException处理异常类型。策略定义有问题吗？为什么不调用断路器后备？

Answer 1

我创建了以下minimum, complete, verifiable example来帮助探讨问题：

注意：不一定是成品；

using Polly;
using Polly.CircuitBreaker;
using System;
using System.Net.Http;
using System.Threading.Tasks;

public class Program
{
    public static void Main()
    {
        int maxRetryCount = 6;
        double circuitBreakDurationSeconds = 0.2 /* experiment with effect of shorter or longer here, eg: change to = 1, and the fallbackForCircuitBreaker is correctly invoked */ ;
        int maxExceptionsBeforeBreaking = 4; /* experiment with effect of fewer here, eg change to = 1, and the fallbackForCircuitBreaker is correctly invoked */
        int maxParallelizations = 2;
        int maxQueuingActions = 2;

        var retryPolicy = Policy.Handle<Exception>(e => (e is HttpRequestException || (/*!(e is BrokenCircuitException) &&*/ e.InnerException is HttpRequestException))) // experiment with introducing the extra (!(e is BrokenCircuitException) && ) clause here, if necessary/desired, depending on goal
            .WaitAndRetry(
                retryCount: maxRetryCount,
                sleepDurationProvider: attempt => TimeSpan.FromMilliseconds(50 * Math.Pow(2, attempt)),
                onRetry: (ex, calculatedWaitDuration, retryCount, context) =>
                {
                    Console.WriteLine(String.Format("Retry => Count: {0}, Wait duration: {1}, Policy Wrap: {2}, Policy: {3}, Endpoint: {4}, Exception: {5}", retryCount, calculatedWaitDuration, context.PolicyWrapKey, context.PolicyKey, context.OperationKey, ex.Message));
                });

        var circuitBreaker = Policy.Handle<Exception>(e => (e is HttpRequestException || e.InnerException is HttpRequestException))
            .CircuitBreaker(maxExceptionsBeforeBreaking,
                TimeSpan.FromSeconds(circuitBreakDurationSeconds),
                onBreak: (ex, breakDuration) => {
                    Console.WriteLine(String.Format("Circuit breaking for {0} ms due to {1}", breakDuration.TotalMilliseconds, ex.Message));
                },
                onReset: () => {
                    Console.WriteLine("Circuit closed again.");
                },
                onHalfOpen: () => { Console.WriteLine("Half open."); });

        var sharedBulkhead = Policy.Bulkhead(maxParallelizations, maxQueuingActions);

        var fallbackForCircuitBreaker = Policy<bool>
             .Handle<BrokenCircuitException>()
            /* .OrInner<BrokenCircuitException>() */ // Consider this if necessary.
            /* .Or<Exception>(e => circuitBreaker.State != CircuitState.Closed) */ // This check will also detect the circuit in anything but healthy state, regardless of the final exception thrown.
             .Fallback(
                 fallbackValue: false,
                 onFallback: (b, context) =>
                 {
                     Console.WriteLine(String.Format("Operation attempted on broken circuit => Policy Wrap: {0}, Policy: {1}, Endpoint: {2}", context.PolicyWrapKey, context.PolicyKey, context.OperationKey));
                 }
             );

        var fallbackForAnyException = Policy<bool>
                .Handle<Exception>()
                .Fallback<bool>(
                    fallbackAction: (context) => { return false; },
                    onFallback: (e, context) =>
                    {
                        Console.WriteLine(String.Format("An unexpected error occured => Policy Wrap: {0}, Policy: {1}, Endpoint: {2}, Exception: {3}", context.PolicyWrapKey, context.PolicyKey, context.OperationKey, e.Exception.Message));
                    }
                );

        var resilienceStrategy = Policy.Wrap(retryPolicy, circuitBreaker, sharedBulkhead);
        var policyWrap = fallbackForAnyException.Wrap(fallbackForCircuitBreaker.Wrap(resilienceStrategy));

        bool outcome = policyWrap.Execute((context) => CallApi("http://www.doesnotexistattimeofwriting.com/"), new Context("some endpoint info"));
    }

    public static bool CallApi(string uri)
    {
        using (var httpClient = new HttpClient() { Timeout = TimeSpan.FromSeconds(1) }) // Consider HttpClient lifetimes and disposal; this pattern is for minimum change from original posted code, not a recommendation.
        {
            Task<HttpResponseMessage> res = httpClient.GetAsync(uri);
            var response = res.Result; // Consider async/await rather than blocking on the returned Task.
            response.EnsureSuccessStatusCode();
            return true;
        }
    }
}

一个以上的因素可能导致fallbackForCircuitBreaker不被调用：

可以将circuitBreakDurationSeconds设置为短于重试之间的各种尝试和等待所花费的总时间。

如果是这样，电路可能会恢复到半开状态。在half-open state或closed state中，按原样重新引发导致电路断开的异常。 BrokenCircuitException仅在（完全）断路阻止呼叫尝试时抛出。

因此，如果在重试耗尽之前电路已恢复为半开状态，则返回到包装回退策略的异常将是HttpRequestException，而不是BrokenCircuitException。

.Handle<Exception>(e => (e is HttpRequestException || e.InnerException is HttpRequestException))子句可能捕获CircuitBreakerException个具有InnerException is HttpRequestException的子句

一个CircuitBreakerException contains the exception which caused the circuit to break as its InnerException。因此，对e.InnerException is HttpRequestException的过度贪婪/较宽松的检查也可能会捕获到CircuitBreakerException具有InnerException is HttpRequestException的情况。这可能取决于您的目标，也可能不是。

我相信对于原始发布的代码，由于其特殊的构造方式，这种情况不会发生。阻塞Task返回的HttpClient.DoSomethingAsync(...)会导致AggregateException->HttpRequestException，这意味着生成的CircuitBreakerException嵌套了两个深的HttpRequestException：

CircuitBreakerException -> AggregateException -> HttpRequestException

因此，这不属于所发布代码中的一个深度检查。但是，请注意，CircuitBreakerException包含一个异常，该异常导致电路以其InnerException的形式断开。这可能导致handle子句仅检查 e.InnerException is HttpRequestException到~~不必要地~~（如果不是您的目标，则出乎意料）重试CircuitBreakerException，如果有：

（a）将代码更改为async / await，这将删除AggregateException，因此导致嵌套仅深一层

（b）将该代码更改为Polly的.HandleInner<HttpRequestException>() syntax，这是递归贪婪的，因此将捕获嵌套的两个深度的CircuitBreakerException->AggregateException->HttpRequestException。

以上代码中的建议/* commented out */ // with additional explanation建议如何调整所发布的代码，以使fallbackForCircuitBreaker能够按预期调用。

另外两个想法：

请考虑尽可能将其更改为async / await。

通过调用HttpClient.DoSomethingAsync()来阻塞.Result可能会影响性能，或者如果与其他异步代码混合使用则可能导致死锁，并且会带来整个AggregateException-with-InnerException痛苦。

考虑HttpClient实例的处置和生存期。

（保留了第3点和第4点，这在其他地方都得到了广泛讨论。）

Polly CircuitBreaker后备广告无法正常运行

1 个答案: