Saga 模式
Saga 最初出现在1987年Hector Garcaa-Molrna & Kenneth Salem发表的一篇名为《Sagas》的论文里。其核心思想是将长事务拆分为多个短事务，借助Saga事务协调器的协调，来保证要么所有操作都成功完成，要么运行相应的补偿事务以撤消先前完成的工作，从而维护多个服务之间的数据一致性。举例而言，假设有个在线购物网站，其后端服务划分为订单服务、支付服务和库存服务。那么一次下订单的Saga流程如下图所示：
在Saga模式中本地事务是Saga 参与者执行的工作单元，每个本地事务都会更新数据库并发布消息或事件以触发 Saga 中的下一个本地事务。如果本地事务失败，Saga 会执行一系列补偿事务，以撤消先前本地事务所做的更改。 对于Saga模式的实现又分为两种形式：
协同式：把Saga 的决策和执行顺序逻辑分布在Saga的每个参与方中，通过交换事件的方式进行流转。示例图如下所示：
编排式：把Saga的决策和执行顺序逻辑集中定义在一个Saga 编排器中。Saga 编排器发出命令式消息给各个Saga 参与方，指示这些参与方执行怎样的操作。
从上图可以看出，对于协同式Saga 存在一个致命的弊端，那就是存在循环依赖的问题，每个Saga参与方都需要订阅所有影响它们的事件，耦合性较高，且由于Saga 逻辑分散在各参与方，不便维护。相对而言，编排式Saga 则实现了关注点分离，协调逻辑集中在编排器中定义，Saga 参与者仅需实现供编排器调用的API 即可。 在.NET 中也有开箱即用的开源框架实现了编排式的Saga事务模型，也就是MassTransit Courier，接下来就来实际探索一番。
MassTransit Courier 简介
MassTransit Courier 是对Routing Slip（路由单） 模式的实现。该模式用于运行时动态指定消息处理步骤，解决不同消息可能有不同消息处理步骤的问题。实现机制是消息处理流程的开始，创建一个路由单，这个路由单定义消息的处理步骤，并附加到消息中，消息按路由单进行传输，每个处理步骤都会查看_路由单_并将消息传递到路由单中指定的下一个处理步骤。 在MassTransit Courier中是通过抽象IActivity和RoutingSlip来实现了Routing Slip模式。通过按需有序组合一系列的Activity，得到一个用来限定消息处理顺序的Routing Slip。而每个Activity的具体抽象就是IActivity和IExecuteActivity。二者的差别在于IActivity定义了Execute和Compensate两个方法，而IExecuteActivitiy仅定义了Execute方法。其中Execute代表正向操作，Compensate代表反向补偿操作。用一个简单的下单流程：创建订单->扣减库存->支付订单举例而言，使用Courier的实现示意图如下所示：
基于Courier 实现编排式Saga事务
那具体如何使用MassTransit Courier来应用编排式Saga 模式呢，接下来就来创建解决方案来实现以上下单流程示例。
创建解决方案
依次创建以下项目，除共享类库项目外，均安装MassTransit和MassTransit.RabbitMQNuGet包。
| 项目 | 项目名 | 项目类型 |
|---|---|---|
| 订单服务 | MassTransit.CourierDemo.OrderService | ASP.NET Core Web API |
| 库存服务 | MassTransit.CourierDemo.InventoryService | Worker Service |
| 支付服务 | MassTransit.CourierDemo.PaymentService | Worker Service |
| 共享类库 | MassTransit.CourierDemo.Shared | Class Library |
三个服务都添加扩展类MassTransitServiceExtensions，并在Program.cs类中调用services.AddMassTransitWithRabbitMq();注册服务。
using System.Reflection;
using MassTransit.CourierDemo.Shared.Models;namespace MassTransit.CourierDemo.InventoryService;public static class MassTransitServiceExtensions
{public static IServiceCollection AddMassTransitWithRabbitMq(this IServiceCollection services){return services.AddMassTransit(x =>{x.SetKebabCaseEndpointNameFormatter();// By default, sagas are in-memory, but should be changed to a durable// saga repository.x.SetInMemorySagaRepositoryProvider();var entryAssembly = Assembly.GetEntryAssembly();x.AddConsumers(entryAssembly);x.AddSagaStateMachines(entryAssembly);x.AddSagas(entryAssembly);x.AddActivities(entryAssembly);x.UsingRabbitMq((context, busConfig) =>{busConfig.Host(host: "localhost",port: 5672,virtualHost: "masstransit",configure: hostConfig =>{hostConfig.Username("guest");hostConfig.Password("guest");});busConfig.ConfigureEndpoints(context);});});}
}订单服务
订单服务作为下单流程的起点，需要承担构建RoutingSlip的职责，因此可以创建一个OrderRoutingSlipBuilder来构建RoutingSlip，代码如下：
using MassTransit.Courier.Contracts;
using MassTransit.CourierDemo.Shared.Models;namespace MassTransit.CourierDemo.OrderService;
public static class OrderRoutingSlipBuilder
{public static RoutingSlip BuildOrderRoutingSlip(CreateOrderDto createOrderDto){var createOrderAddress = new Uri("queue:create-order_execute");var deduceStockAddress = new Uri("queue:deduce-stock_execute");var payAddress = new Uri("queue:pay-order_execute"); var routingSlipBuilder = new RoutingSlipBuilder(Guid.NewGuid());routingSlipBuilder.AddActivity(name: "order-activity",executeAddress: createOrderAddress,arguments: createOrderDto);routingSlipBuilder.AddActivity(name: "deduce-stock-activity", executeAddress: deduceStockAddress);routingSlipBuilder.AddActivity(name: "pay-activity", executeAddress: payAddress);var routingSlip = routingSlipBuilder.Build();return routingSlip;}
}从以上代码可知，构建一个路由单需要以下几步：
明确业务用例涉及的具体用例，本例中为：
创建订单：CreateOrder
扣减库存：DeduceStock
支付订单：PayOrder
根据用例名，按短横线隔开命名法（kebab-case）定义用例执行地址，格式为
queue:<usecase>_execute，本例中为：创建订单执行地址：queue:create-order_execute
创建订单执行地址：queue:deduce-stock_execute
创建订单执行地址：queue:pay-order_execute
创建路由单：
通过
RoutingSlipBuilder(Guid.NewGuid())创建路由单构建器实例根据业务用例流转顺序，调用
AddActivity()方法依次添加Activity用来执行用例，因为第一个创建订单用例需要入口参数，因此传入了一个CreateOrderDtoDTO（Data Transfer Object）对象调用
Build()方法创建路由单
对于本例而言，由于下单流程是固定流程，因此以上路由单的构建也是按业务用例进行定义的。而路由单的强大之处在于，可以按需动态组装。在实际电商场景中，有些订单是无需执行库存扣减的，比如充值订单，对于这种情况，仅需在创建路由单时判断若为充值订单则不添加扣减库存的Activity即可。 对于订单服务必然要承担创建订单的职责，定义CreateOrderActivity（Activity的命名要与上面定义的用例对应）如下，其中OrderRepository为一个静态订单仓储类：
public class CreateOrderActivity : IActivity<CreateOrderDto, CreateOrderLog>
{private readonly ILogger<CreateOrderActivity> _logger;public CreateOrderActivity(ILogger<CreateOrderActivity> logger){_logger = logger;}// 订单创建public async Task<ExecutionResult> Execute(ExecuteContext<CreateOrderDto> context){var order = await CreateOrder(context.Arguments);var log = new CreateOrderLog(order.OrderId, order.CreatedTime);_logger.LogInformation($"Order [{order.OrderId}] created successfully!");return context.CompletedWithVariables(log, new {order.OrderId});}private async Task<Order> CreateOrder(CreateOrderDto orderDto){var shoppingItems =orderDto.ShoppingCartItems.Select(item => new ShoppingCartItem(item.SkuId, item.Price, item.Qty));var order = new Order(orderDto.CustomerId).NewOrder(shoppingItems.ToArray());await OrderRepository.Insert(order);return order;}// 订单补偿（取消订单）public async Task<CompensationResult> Compensate(CompensateContext<CreateOrderLog> context){var order = await OrderRepository.Get(context.Log.OrderId);order.CancelOrder();var exception = context.Message.ActivityExceptions.FirstOrDefault();_logger.LogWarning($"Order [{order.OrderId} has been canceled duo to {exception.ExceptionInfo.Message}!");return context.Compensated();}
}从以上代码可知，实现一个Activity，需要以下步骤：
定义实现
IActivity<in TArguments, in TLog>需要的参数类：TArguments对应正向执行入口参数，会在Execute方法中使用，本例中为CreateOrderDto，用于订单创建。TLog对应反向补偿参数，会在Compensate方法中使用，本例中为CreateOrderLog，用于订单取消。
实现
IActivity<in TArguments, in TLog>接口中的Execute方法：具体用例的实现，本例中对应订单创建逻辑
创建
TLog反向补偿参数实例，以便业务异常时能够按需补偿返回Activity执行结果，并按需传递参数至下一个Activity，本例仅传递订单Id至下一流程。
实现
IActivity<in TArguments, in TLog>接口中的Compensate方法：具体反向补偿逻辑的实现，本例中对应取消订单
返回反向补偿执行结果
订单服务的最后一步就是定义WebApi来接收创建订单请求，为简要起便创建OrderController如下：
using MassTransit.CourierDemo.Shared.Models;
using Microsoft.AspNetCore.Mvc;namespace MassTransit.CourierDemo.OrderService.Controllers;[ApiController]
[Route("[controller]")]
public class OrderController : ControllerBase
{private readonly IBus _bus;public OrderController(IBus bus){_bus = bus;}[HttpPost]public async Task<IActionResult> CreateOrder(CreateOrderDto createOrderDto){// 创建订单路由单var orderRoutingSlip = OrderRoutingSlipBuilder.BuildOrderRoutingSlip(createOrderDto);// 执行订单流程await _bus.Execute(orderRoutingSlip);return Ok();}
}库存服务
库存服务在整个下单流程的职责主要是库存的扣减和返还，但由于从上游用例仅传递了OrderId参数到库存扣减Activity，因此在库存服务需要根据OrderId 去请求订单服务获取要扣减的库存项才能执行扣减逻辑。而这可以通过使用MassTransit的Reqeust/Response 模式来实现，具体步骤如下：
在共享类库
MassTransit.CourierDemo.Shared中定义IOrderItemsRequest和IOrderItemsResponse：
namespace MassTransit.CourierDemo.Shared.Models;public interface IOrderItemsRequest
{public string OrderId { get; }
}
public interface IOrderItemsResponse
{public List<DeduceStockItem> DeduceStockItems { get; set; }public string OrderId { get; set; }
}在订单服务中实现
IConsumer<IOrderItemsRequest:
using MassTransit.CourierDemo.OrderService.Repositories;
using MassTransit.CourierDemo.Shared.Models;namespace MassTransit.CourierDemo.OrderService.Consumers;public class OrderItemsRequestConsumer : IConsumer<IOrderItemsRequest>
{public async Task Consume(ConsumeContext<IOrderItemsRequest> context){var order = await OrderRepository.Get(context.Message.OrderId);await context.RespondAsync<IOrderItemsResponse>(new{order.OrderId, DeduceStockItems = order.OrderItems.Select(item => new DeduceStockItem(item.SkuId, item.Qty)).ToList()});}
}在库存服务注册
service.AddMassTransit()中注册x.AddRequestClient<IOrderItemsRequest>();：
using System.Reflection;
using MassTransit.CourierDemo.Shared.Models;namespace MassTransit.CourierDemo.InventoryService;public static class MassTransitServiceExtensions
{public static IServiceCollection AddMassTransitWithRabbitMq(this IServiceCollection services){return services.AddMassTransit(x =>{//... x.AddRequestClient<IOrderItemsRequest>();//...});}
}在需要的类中注册
IRequestClient<OrderItemsRequest>服务即可。
最终扣减库存的Activity实现如下：
public class DeduceStockActivity : IActivity<DeduceOrderStockDto, DeduceStockLog>
{private readonly IRequestClient<IOrderItemsRequest> _orderItemsRequestClient;private readonly ILogger<DeduceStockActivity> _logger;public DeduceStockActivity(IRequestClient<IOrderItemsRequest> orderItemsRequestClient,ILogger<DeduceStockActivity> logger){_orderItemsRequestClient = orderItemsRequestClient;_logger = logger;}// 库存扣减public async Task<ExecutionResult> Execute(ExecuteContext<DeduceOrderStockDto> context){var deduceStockDto = context.Arguments;var orderResponse =await _orderItemsRequestClient.GetResponse<IOrderItemsResponse>(new { deduceStockDto.OrderId });if (!CheckStock(orderResponse.Message.DeduceStockItems))return context.Faulted(new Exception("insufficient stock"));DeduceStocks(orderResponse.Message.DeduceStockItems);var log = new DeduceStockLog(deduceStockDto.OrderId, orderResponse.Message.DeduceStockItems);_logger.LogInformation($"Inventory has been deducted for order [{deduceStockDto.OrderId}]!");return context.CompletedWithVariables(log, new { log.OrderId });}// 库存检查private bool CheckStock(List<DeduceStockItem> deduceItems){foreach (var stockItem in deduceItems){if (InventoryRepository.GetStock(stockItem.SkuId) < stockItem.Qty) return false;}return true;}private void DeduceStocks(List<DeduceStockItem> deduceItems){foreach (var stockItem in deduceItems){InventoryRepository.TryDeduceStock(stockItem.SkuId, stockItem.Qty);}}//库存补偿public Task<CompensationResult> Compensate(CompensateContext<DeduceStockLog> context){foreach (var deduceStockItem in context.Log.DeduceStockItems){InventoryRepository.ReturnStock(deduceStockItem.SkuId, deduceStockItem.Qty);}_logger.LogWarning($"Inventory has been returned for order [{context.Log.OrderId}]!");return Task.FromResult(context.Compensated());}
}支付服务
对于下单流程的支付用例来说，要么成功要么失败，并不需要像以上两个服务一样定义补偿逻辑，因此仅需要实现IExecuteActivity<in TArguments>接口即可，该接口仅定义了Execute接口方法，具体PayOrderActivity实现如下：
using MassTransit.CourierDemo.Shared;
using MassTransit.CourierDemo.Shared.Models;namespace MassTransit.CourierDemo.PaymentService.Activities;public class PayOrderActivity : IExecuteActivity<PayDto>
{private readonly IBus _bus;private readonly IRequestClient<IOrderAmountRequest> _client;private readonly ILogger<PayOrderActivity> _logger;public PayOrderActivity(IBus bus,IRequestClient<IOrderAmountRequest> client,ILogger<PayOrderActivity> logger){_bus = bus;_client = client;_logger = logger;}public async Task<ExecutionResult> Execute(ExecuteContext<PayDto> context){var response = await _client.GetResponse<IOrderAmountResponse>(new { context.Arguments.OrderId }); // do payment...if (response.Message.Amount % 2 == 0){_logger.LogInformation($"Order [{context.Arguments.OrderId}] paid successfully!");return context.Completed();}_logger.LogWarning($"Order [{context.Arguments.OrderId}] payment failed!");return context.Faulted(new Exception("Order payment failed due to insufficient account balance."));}
}以上代码中也使用了MassTransit的Reqeust/Response 模式来获取订单要支付的余额，并根据订单金额是否为偶数来模拟支付失败。
运行结果
启动三个项目，并在Swagger中发起订单创建请求，如下图所示：
由于订单总额为奇数，因此支付会失败，最终控制台输出如下图所示：
打开RabbitMQ后台，可以看见MassTransit按照约定创建了以下队列用于服务间的消息传递：
但你肯定好奇本文中使用的路由单具体是怎样实现的？简单，停掉库存服务，再发送一个订单创建请求，然后从队列获取未消费的消息即可解开谜底。以下是抓取的一条消息示例：
{"messageId": "ac5d0000-e330-482a-b7bc-08dada7915ab","requestId": null,"correlationId": "ce8af31b-a65c-4dfa-915c-4ae5174820f9","conversationId": "ac5d0000-e330-482a-28a5-08dada7915ad","initiatorId": null,"sourceAddress": "rabbitmq://localhost/masstransit/THINKPAD_MassTransitCourierDemoOrderService_bus_itqoyy8dgbrniyeobdppw6engn?temporary=true","destinationAddress": "rabbitmq://localhost/masstransit/deduce-stock_execute?bind=true","responseAddress": null,"faultAddress": null,"messageType": ["urn:message:MassTransit.Courier.Contracts:RoutingSlip"],"message": {"trackingNumber": "ce8af31b-a65c-4dfa-915c-4ae5174820f9","createTimestamp": "2022-12-10T06:38:01.5452768Z","itinerary": [{"name": "deduce-stock-activity","address": "queue:deduce-stock_execute","arguments": {}},{"name": "pay-activity","address": "queue:pay-order_execute","arguments": {}}],"activityLogs": [{"executionId": "ac5d0000-e330-482a-7cb2-08dada7915bf","name": "order-activity","timestamp": "2022-12-10T06:38:01.7115314Z","duration": "00:00:00.0183136","host": {"machineName": "THINKPAD","processName": "MassTransit.CourierDemo.OrderService","processId": 23980,"assembly": "MassTransit.CourierDemo.OrderService","assemblyVersion": "1.0.0.0","frameworkVersion": "6.0.9","massTransitVersion": "8.0.7.0","operatingSystemVersion": "Microsoft Windows NT 10.0.19044.0"}}],"compensateLogs": [{"executionId": "ac5d0000-e330-482a-7cb2-08dada7915bf","address": "rabbitmq://localhost/masstransit/create-order_compensate","data": {"orderId": "8c47a1db-cde3-43bb-a809-644f36e7ca99","createdTime": "2022-12-10T14:38:01.7272895+08:00"}}],"variables": {"orderId": "8c47a1db-cde3-43bb-a809-644f36e7ca99"},"activityExceptions": [],"subscriptions": []},"expirationTime": null,"sentTime": "2022-12-10T06:38:01.774618Z","headers": {"MT-Forwarder-Address": "rabbitmq://localhost/masstransit/create-order_execute"}
}从中可以看到信封中的message.itinerary定义了消息的行程，从而确保消息按照定义的流程进行流转。同时通过message.compensateLogs来指引若失败将如何回滚。
总结
通过以上示例的讲解，相信了解到MassTransit Courier的强大之处。Courier中的RoutingSlip充当着事务编排器的角色，将Saga的决策和执行顺序逻辑封装在消息体内随着消息进行流转，从而确保各服务仅需关注自己的业务逻辑，而无需关心事务的流转，真正实现了关注点分离。
