微服务简单实现最终一致性

有花时间去研究masstransit的saga,英文水平不过关,始终无法实现上手他的代码编排的业务,遗憾。

本文通过rabbit和sqlserver实现下单,更新库存,更新产品,模拟数据最终一致性。

 

项目结构如下,reportService可有可无,这里就相当一个链条,只要两节走通了后面可以接龙,本文有用到不省略。流程:orderservice=>eComm=>reportservice 。

 

下面先看看order的配置,通过控制器新增订单同时发布订单信息到order_exchange交换机,Key是”order.created,这样就把订单推送到了队列,等到库存服务获取订单去更新库存。

  // POST api/<OrderController>
        [HttpPost]
        public async Task Post([FromBody] OrderDetail orderDetail)
        {
            var id = await orderCreator.Create(orderDetail);
            publisher.Publish(JsonConvert.SerializeObject(new OrderRequest { 
                OrderId = id,
                ProductId = orderDetail.ProductId,
                Quantity = orderDetail.Quantity


            }), "order.created", null);
        } 

更新库存的代码,然后再发送消息告诉order服务,这里有哪个try包裹,如果这里有失败会触发catch,发送减库存失败的消息。order服务消费到这条消息就会执行相应的删除订单操作。代码如下:

using Ecomm.DataAccess;
using Ecomm.Models;
using Microsoft.Extensions.Hosting;
using Newtonsoft.Json;
using Plain.RabbitMQ;
using System;
using System.Collections.Generic;
using System.Linq;
using System.Threading;
using System.Threading.Tasks;

namespace Ecomm
{
    public class OrderCreatedListener : IHostedService
    {
        private readonly IPublisher publisher;
        private readonly ISubscriber subscriber;
        private readonly IInventoryUpdator inventoryUpdator;

        public OrderCreatedListener(IPublisher publisher, ISubscriber subscriber, IInventoryUpdator inventoryUpdator)
        {
            this.publisher = publisher;
            this.subscriber = subscriber;
            this.inventoryUpdator = inventoryUpdator;
        }

        public Task StartAsync(CancellationToken cancellationToken)
        {
            subscriber.Subscribe(Subscribe);
            return Task.CompletedTask;
        }

        private bool Subscribe(string message, IDictionary<string, object> header)
        {
            var response = JsonConvert.DeserializeObject<OrderRequest>(message);
            try
            {
                inventoryUpdator.Update(response.ProductId, response.Quantity).GetAwaiter().GetResult();
                publisher.Publish(JsonConvert.SerializeObject(
                    new InventoryResponse { OrderId = response.OrderId, IsSuccess = true }
                    ), "inventory.response", null);
            }
            catch (Exception)
            {
                publisher.Publish(JsonConvert.SerializeObject(
                    new InventoryResponse { OrderId = response.OrderId, IsSuccess = false }
                    ), "inventory.response", null);
            }

            return true;
        }

        public Task StopAsync(CancellationToken cancellationToken)
        {
            return Task.CompletedTask;
        }
    }
}

  

using Ecomm.Models;
using Microsoft.Extensions.Hosting;
using Newtonsoft.Json;
using Plain.RabbitMQ;
using System.Collections.Generic;
using System.Threading;
using System.Threading.Tasks;

namespace OrderService
{
    public class InventoryResponseListener : IHostedService
    {
        private readonly ISubscriber subscriber;
        private readonly IOrderDeletor orderDeletor;

        public InventoryResponseListener(ISubscriber subscriber, IOrderDeletor orderDeletor)
        {
            this.subscriber = subscriber;
            this.orderDeletor = orderDeletor;
        }

        public Task StartAsync(CancellationToken cancellationToken)
        {
            subscriber.Subscribe(Subscribe);
            return Task.CompletedTask;
        }

        private bool Subscribe(string message, IDictionary<string, object> header)
        {
            var response = JsonConvert.DeserializeObject<InventoryResponse>(message);
            if (!response.IsSuccess)
            {
                orderDeletor.Delete(response.OrderId).GetAwaiter().GetResult();
            }
            return true;
        } 

        public Task StopAsync(CancellationToken cancellationToken)
        {
            return Task.CompletedTask;
        }
    }
}

  上面的代码是整个服务的核心业务,也很简单就是队列相互通信相互确认操作是否顺利,失败就执行回归操作,而这里我们都会写好对应补偿代码:

using Dapper;
using System;
using System.Collections.Generic;
using System.Data.SqlClient;
using System.Linq;
using System.Threading.Tasks;

namespace OrderService
{
    public class OrderDeletor : IOrderDeletor
    {
        private readonly string connectionString;

        public OrderDeletor(string connectionString)
        {
            this.connectionString = connectionString;
        }

        public async Task Delete(int orderId)
        {
            using var connection = new SqlConnection(connectionString);
            connection.Open();
            using var transaction = connection.BeginTransaction();
            try
            {
                await connection.ExecuteAsync("DELETE FROM OrderDetail WHERE OrderId = @orderId", new { orderId }, transaction: transaction);
                await connection.ExecuteAsync("DELETE FROM [Order] WHERE Id = @orderId", new { orderId }, transaction: transaction);
                transaction.Commit();
            }
            catch
            {
                transaction.Rollback();
            }
        }
    }
}

库存服务里有发布产品的接口,这里没有做过多的处理,只是把产品新增放到队列,供后面的ReportService服务获取,该服务拿到后会执行产品数量扣除:

using Microsoft.Extensions.Hosting;
using Newtonsoft.Json;
using Plain.RabbitMQ;
using System.Collections.Generic;
using System.Linq;
using System.Threading;
using System.Threading.Tasks;

namespace ReportService
{
    public class ReportDataCollector : IHostedService
    {
        private const int DEFAULT_QUANTITY = 100;
        private readonly ISubscriber subscriber;
        private readonly IMemoryReportStorage memoryReportStorage;

        public ReportDataCollector(ISubscriber subscriber, IMemoryReportStorage memoryReportStorage)
        {
            this.subscriber = subscriber;
            this.memoryReportStorage = memoryReportStorage;
        }

        public Task StartAsync(CancellationToken cancellationToken)
        {
            subscriber.Subscribe(Subscribe);
            return Task.CompletedTask;
        }
        private bool Subscribe(string message, IDictionary<string, object> header)
        //private bool ProcessMessage(string message, IDictionary<string, object> header)
        {
            if (message.Contains("Product"))
            {
                var product = JsonConvert.DeserializeObject<Product>(message);
                if (memoryReportStorage.Get().Any(r => r.ProductName == product.ProductName))
                {
                    return true;
                }
                else
                {
                    memoryReportStorage.Add(new Report
                    {
                        ProductName = product.ProductName,
                        Count = DEFAULT_QUANTITY
                    });
                }
            }
            else
            {
                var order = JsonConvert.DeserializeObject<Order>(message);
                if(memoryReportStorage.Get().Any(r => r.ProductName == order.Name))
                {
                    memoryReportStorage.Get().First(r => r.ProductName == order.Name).Count -= order.Quantity;
                }
                else
                {
                    memoryReportStorage.Add(new Report
                    {
                        ProductName = order.Name,
                        Count = DEFAULT_QUANTITY - order.Quantity
                    });
                }
            }
            return true;
        }

        public Task StopAsync(CancellationToken cancellationToken)
        {
            return Task.CompletedTask;
        }
    }
}

到这里整个流程大概如此。只要理清楚了订单和库存更新这里的业务,后面差不多一样,可以无限递归。代码文末有链接供下载。

这里有一个地方的代码如下,新增库存的时候同时发布消息。假如新增完订单后面崩掉了,这里是个原子操作最佳。

        [HttpPost]
        public async Task Post([FromBody] OrderDetail orderDetail)
        {
            var id = await orderCreator.Create(orderDetail);
            publisher.Publish(JsonConvert.SerializeObject(new OrderRequest { 
                OrderId = id,
                ProductId = orderDetail.ProductId,
                Quantity = orderDetail.Quantity


            }), "order.created", null);
        }

  很遗憾masstransit的saga还没有整明白,那就上cap,完成业务一致性。加了点cap代码因为之前是dapper,所以加了dbcontext和cap相关代码有点小乱。核心代码如下:

 

using DotNetCore.CAP;
using MediatR;
using OrderService.Command;
using System.Threading;
using Ecomm.Models;
using System.Collections.Generic;

namespace OrderService.Handler
{
    public class InsertOrderDetailHandler : IRequestHandler<InsertOrderDetailCommand, InsertOrderDetailModel>
    {
        private readonly OrderDbContext context;
        private readonly ICapPublisher cap;
        public InsertOrderDetailHandler(OrderDbContext context, ICapPublisher cap)
        {
            this.context = context;
            this.cap = cap;
        }
        public async System.Threading.Tasks.Task<InsertOrderDetailModel> Handle(InsertOrderDetailCommand request, CancellationToken cancellationToken)
        {
            using(var trans =context.Database.BeginTransaction(cap))
            {
                var order =  context.Orders.Add(new Order
                {
                    UpdatedTime = System.DateTime.Today,
                    UserId = request.UserId,
                    UserName = request.UserName
                });
                var orderDetail = context.OrderDetails.Add(new OrderDetail
                {
                    OrderId = order.Entity.Id,
                    ProductId = request.ProductId,
                    Quantity = request.Quantity,
                    ProductName = request.ProductName,
                });
                 context.SaveChanges();
                
                cap.Publish<OrderRequest>("order.created", new OrderRequest
                {
                    OrderId = order.Entity.Id,
                    ProductId = orderDetail.Entity.ProductId,
                    Quantity = orderDetail.Entity.Quantity
                }, new Dictionary<string,string>()) ;
                 trans.Commit();
                return new InsertOrderDetailModel { OrderDetailid = orderDetail.Entity.Id, OrderId = order.Entity.Id, Success = true };
            }
        }
    }
}

 到这里差不多要结束了,这里的代码都可以调试运行的。因为加了cap,order服务有两套rabbitmq的配置,有冗余,而且有点坑。调试的时候注意,Plain.RabbitMQ支持的交换机不是持久化的,而cap是持久化的,所以有点不兼容。第一次运行可以先确保Plain.RabbitMQ正常,再删掉交换机,cap跑起来了再建持久化交换机,这样cap消息就会被rabbitmq接收,后面就会被库存服务消费。因为我这里cap不会自动绑定队列,Plain.RabbitMQ是可以的。所以需要新建交换机后再绑定队列。而且这里队列以Plain.RabbitMQ生成的名字来绑定。要不然又可能会调试踩坑无法出坑。 用cap不注意你连消息队列都看不到,看到了队列也看不到消费数据,这点不知道是我不会还是cap有什么难的配置。结束。。。

上例项目demo:

liuzhixin405/SimpleOrders_Next (github.com)

超简单微服务demo

liuzhixin405/SimpleOrders (github.com)