执行块的最长时间............带有重试和中间延迟

Posted

技术标签:

【中文标题】执行块的最长时间............带有重试和中间延迟【英文标题】:Maximum Time for an execution block.........with retry and in-between-delay 【发布时间】:2017-11-29 19:47:18 【问题描述】:

我正在尝试编写策略。

我希望我的执行代码运行最长时间(在示例中为 10 秒)。 但我也想重试 x 次(样本中为 3 次)。并在失败之​​间暂停(示例中为 2 秒)。

我已经操纵我的存储过程来人为地延迟测试我的行为。

按照编码(如下代码),我的数据集在 30 秒后填充(仅供参考:30 秒是存储过程中的硬编码值)。所以我的执行代码在 10 秒后没有退出......

理想情况下,我看到代码在前两次尝试 10 秒后退出,然后在进行第三次尝试(因为存储过程不会人为延迟)。显然这不是真正的代码,但奇怪的存储过程给了我一种测试行为的方法。

我的存储过程:

USE [Northwind]
GO


/* CREATE */ ALTER PROCEDURE [dbo].[uspWaitAndReturn]
(
    @InvokeDelay bit
)

AS

SET NOCOUNT ON;

if ( @InvokeDelay > 0)
BEGIN
    WAITFOR DELAY '00:00:30';  
END

select top 1 * from dbo.Customers c order by newid()

GO

我的 C#/Polly/数据库代码:

    public DataSet GetGenericDataSet()
    
        DataSet returnDs = null;

        int maxRetryAttempts = 3; /* retry attempts */
        TimeSpan pauseBetweenFailuresTimeSpan = TimeSpan.FromSeconds(2); /* pause in between failures */
        Policy timeoutAfter10SecondsPolicy = Policy.Timeout(TimeSpan.FromSeconds(10)); /* MAGIC SETTING here, my code inside the below .Execute block below would bail out after 10 seconds */
        Policy retryThreeTimesWith2SecondsInBetweenPolicy = Policy.Handle<Exception>().WaitAndRetry(maxRetryAttempts, i => pauseBetweenFailuresTimeSpan);
        Policy aggregatePolicy = timeoutAfter10SecondsPolicy.Wrap(retryThreeTimesWith2SecondsInBetweenPolicy);

        int attemptCounter = 0; /* used to track the attempt and conditionally set the @InvokeDelay value for the stored procedure */

        aggregatePolicy.Execute(() =>
        
            try
            
                attemptCounter++;

                /* Microsoft.Practices.EnterpriseLibrary.Data code */
                ////////DatabaseProviderFactory factory = new DatabaseProviderFactory();
                ////////Database db = factory.CreateDefault();
                ////////DbCommand dbc = db.GetStoredProcCommand("dbo.uspWaitAndReturn");
                ////////dbc.CommandTimeout = 120;
                ////////db.AddInParameter(dbc, "@InvokeDelay", DbType.Boolean, attemptCounter < maxRetryAttempts ? true : false); /* if i'm not on my last attempt, then pass in true to cause the artificial delay */
                ////////DataSet ds;
                ////////ds = db.ExecuteDataSet(dbc);
                ////////returnDs = ds;

                using (SqlConnection conn = new SqlConnection(@"MyConnectionStringValueHere"))
                
                    SqlCommand sqlComm = new SqlCommand("[dbo].[uspWaitAndReturn]", conn);
                    sqlComm.Parameters.AddWithValue("@InvokeDelay", attemptCounter < maxRetryAttempts ? true : false);
                    sqlComm.CommandType = CommandType.StoredProcedure;

                    SqlDataAdapter da = new SqlDataAdapter();
                    da.SelectCommand = sqlComm;
                    DataSet ds = new DataSet();
                    da.Fill(ds);
                    returnDs = ds;
                

            
            catch (Exception ex)
            
                string temp = ex.Message;
                throw;
            
        );

        return returnDs;
    

使用语句:

using System;
using System.Data;
using System.Data.Common;
using System.Data.SqlClient;
////    using Microsoft.Practices.EnterpriseLibrary.Data;
using Polly;

版本(packages.config)

<?xml version="1.0" encoding="utf-8"?>
<packages>
  <package id="CommonServiceLocator" version="1.0" targetFramework="net40" />
  <package id="EnterpriseLibrary.Common" version="6.0.1304.0" targetFramework="net45" />
  <package id="EnterpriseLibrary.Data" version="6.0.1304.0" targetFramework="net45" />


  <package id="Polly" version="5.6.1" targetFramework="net45" />


/>
</packages>

追加:

在@mountain traveller 的精彩回答之后,我有一个工作示例:

重点是:

添加了 TimeoutStrategy.Pessimistic

并添加了 DbCommand.Cancel() 调用(如果您不使用企业库,则添加 SqlCommand.Cancel())来终止(以前的)命令,否则它们将继续运行(不好)。

我还必须“反转”我的“Policy aggregatePolicy”。

    public DataSet GetGenericDataSet()
    
        DataSet returnDs = null;

        DbCommand dbc = null; /* increase scope so it can be cancelled */

        int maxRetryAttempts = 3; /* retry attempts */
        TimeSpan pauseBetweenFailuresTimeSpan = TimeSpan.FromSeconds(2); /* pause in between failures */
        Policy timeoutAfter10SecondsPolicy = Policy.Timeout(
            TimeSpan.FromSeconds(10), 
            TimeoutStrategy.Pessimistic,
            (context, timespan, task) =>
            
                string x = timespan.Seconds.ToString();
                if (null != dbc)
                
                    dbc.Cancel();
                    dbc = null;
                
            );

        Policy retryThreeTimesWith2SecondsInBetweenPolicy = Policy.Handle<Exception>().WaitAndRetry(maxRetryAttempts, i => pauseBetweenFailuresTimeSpan);
        ////Policy aggregatePolicy = timeoutAfter10SecondsPolicy.Wrap(retryThreeTimesWith2SecondsInBetweenPolicy);
        Policy aggregatePolicy = retryThreeTimesWith2SecondsInBetweenPolicy.Wrap(timeoutAfter10SecondsPolicy);

        int attemptCounter = 0; /* used to track the attempt and conditionally set the @InvokeDelay value for the stored procedure */

        aggregatePolicy.Execute(() =>
        
            try
            
                attemptCounter++;

                /* Microsoft.Practices.EnterpriseLibrary.Data code */
                DatabaseProviderFactory factory = new DatabaseProviderFactory();
                Database db = factory.CreateDefault();
                dbc = db.GetStoredProcCommand("dbo.uspWaitAndReturn");
                dbc.CommandTimeout = 120;
                db.AddInParameter(dbc, "@InvokeDelay", DbType.Boolean, attemptCounter < maxRetryAttempts ? true : false); /* if i'm not on my last attempt, then pass in true to cause the artificial delay */
                DataSet ds;
                ds = db.ExecuteDataSet(dbc);
                returnDs = ds;

                ////////using (SqlConnection conn = new SqlConnection(@"YOUR_VALUE_HERE"))
                ////////
                ////////    SqlCommand sqlComm = new SqlCommand("[dbo].[uspWaitAndReturn]", conn);
                ////////    sqlComm.Parameters.AddWithValue("@InvokeDelay", attemptCounter < maxRetryAttempts ? true : false);
                ////////    sqlComm.CommandType = CommandType.StoredProcedure;

                ////////    SqlDataAdapter da = new SqlDataAdapter();
                ////////    da.SelectCommand = sqlComm;
                ////////    DataSet ds = new DataSet();
                ////////    da.Fill(ds);
                ////////    returnDs = ds;
                ////////
            
            catch (SqlException sqlex)
            
                switch (sqlex.ErrorCode)
                
                    case -2146232060:
                        /* I couldn't find a more concrete way to find this specific exception, -2146232060 seems to represent alot of things */
                        if (!sqlex.Message.Contains("cancelled"))
                        
                            throw;
                        

                        break;
                    default:
                        throw;
                
            
        );

        return returnDs;
    

【问题讨论】:

【参考方案1】:

波莉TimeoutPolicycomes in two modes:

Optimistic Timeout 通过CancellationToken 通过co-operative cancellation 强制超时。它期望执行的代表响应CancellationToken

Pessimistic Timeout 对响应CancellationToken的委托强制超时

您正在执行的委托不尊重任何CancellationToken。因此(对于发布的原始代码)您需要将策略配置为使用TimeoutStrategy.Pessimistic

Policy timeoutAfter10SecondsPolicy = Policy.Timeout(TimeSpan.FromSeconds(10), TimeoutStrategy.Pessimistic);

(在发布的原始代码中,Policy timeoutAfter10SecondsPolicy = Policy.Timeout(TimeSpan.FromSeconds(10)); 采用了TimeoutStrategy.Optimistic,因为这是默认设置。)


以上是为了解释为什么您没有看到TimeoutPolicy 在提供的代码中工作。 :注意discussion in the Polly wiki about what pessimistic timeout means:它允许调用线程离开等待执行的委托,但不会取消线程/Task 运行该委托代表。所以使用中的 SQL 资源不会在超时时被释放。

为确保 SQL 资源在超时时释放,您需要扩展代码以在超时时取消 SQL 操作。您可以使用SqlCommand.Cancel() method,例如this *** answer 中所示。 Polly 的TimeoutPolicy can be configured with an onTimeout delegate 发生超时时调用:即配置此委托以调用适当的SqlCommand.Cancel()

或者,可以将整个代码移至异步方法,并使用SqlCommand.ExecuteReaderAsync(CancellationToken) 之类的东西,再加上由CancellationToken 驱动的Polly 的optimistic timeout。但这是一个更广泛的讨论。

【讨论】:

在你写这篇文章的时候,我正在处理 DbCommand.Cancel() (当然具体来说是 SqlCommand.Cancel() )。谢谢。我想我的 aggregatePolicy 也“向后”。 我在原始问题中附加了完整的工作代码和 Sql Profiler 屏幕截图。再次感谢。

以上是关于执行块的最长时间............带有重试和中间延迟的主要内容,如果未能解决你的问题,请参考以下文章

dubbo超时重试和异常处理

JPA 事务回滚重试和恢复:将实体与自动递增的@Version 合并

Spring Boot Cloud + Ribbon + Feign + Hystrix + Zookeeper:重试和失败是怎么回事?

AWS 中的错误重试和指数退避

AWS 中的错误重试和指数退避 Error Retries and Exponential Backoff in AWS

异步通信rabbitmq——消息重试