为什么在Azure中进行扩展时不会调用Stopping / OnStop？

Question

当我减少角色实例计数时， RoleEnvironment.Stopping / RoleEntryPoint.OnStop（）为死亡实例调用 。重启实例或停止部署时，会调用 。我做错了什么，或者我不应该在这种情况下需要清理？

我有一个简单的辅助角色（VS2012更新1，默认的云项目，一个辅助角色，添加了smarx's Table storage trace listener）。这里所有代码;没有其他依赖：

using System;
using System.Collections.Generic;
using System.Data.Services.Client;
using System.Diagnostics;
using System.Linq;
using System.Net;
using System.Threading;
using Microsoft.WindowsAzure;
using Microsoft.WindowsAzure.Diagnostics;
using Microsoft.WindowsAzure.ServiceRuntime;
using Microsoft.WindowsAzure.StorageClient;

namespace WorkerRole1
{
    public class WorkerRole : RoleEntryPoint
    {
        bool shouldRun = true;
        EventWaitHandle runFinished = new EventWaitHandle(true, EventResetMode.ManualReset);

        public override bool OnStart()
        {
            ServicePointManager.DefaultConnectionLimit = 12;
            RoleEnvironment.Stopping += (object sender, RoleEnvironmentStoppingEventArgs e) => {
                Trace.WriteLine("WorkerRole1 Stopping called", "Information");
                shouldRun = false;
            };
            return base.OnStart();
        }

        public override void Run()
        {
            runFinished.Reset();
            try {
                Trace.WriteLine("WorkerRole1 entry point called", "Information");
                while (shouldRun) {
                    Thread.Sleep(10000);
                    Trace.WriteLine("Working", "Information");
                }
                Trace.WriteLine("Finished", "Information");
            } finally {
                runFinished.Set();
            }
        }

        public override void OnStop()
        {
            Trace.WriteLine("OnStop: Waiting for Run() to finish", "Information");
            runFinished.WaitOne();
            Trace.WriteLine("OnStop: Run() finished", "Information");
            base.OnStop();
        }
    }

    public class LogMessage : TableServiceEntity
    {
        public DateTime Time { get; set; }
        public string Message { get; set; }
        public string InstanceId { get; set; }
        public string Category { get; set; }

        public LogMessage() { }
        public LogMessage(string message, string category)
        {
            Message = message;
            Category = category;
            Time = DateTime.UtcNow;
            InstanceId = RoleEnvironment.CurrentRoleInstance.Id;
            PartitionKey = RoleEnvironment.DeploymentId;
            RowKey = (DateTime.MaxValue.Ticks - Time.Ticks).ToString("d19");
        }
    }

    public class TableTraceListener : TraceListener
    {
        private TableServiceContext _context = null;
        private TableServiceContext context
        {
            get
            {
                if (_context == null) {
                    var tables = CloudStorageAccount
                        .Parse(RoleEnvironment.GetConfigurationSettingValue(
                            Attributes["connectionStringName"] ?? "DataConnectionString"))
                        .CreateCloudTableClient();
                    tables.CreateTableIfNotExist("log");
                    _context = tables.GetDataServiceContext();
                    _context.MergeOption = MergeOption.NoTracking;
                }
                return _context;
            }
        }

        protected override string[] GetSupportedAttributes() { return new[] { "connectionStringName" }; }

        public override void Write(string message, string category)
        {
            context.AddObject("log", new LogMessage(message, category));
            context.SaveChangesWithRetries();
        }

        public override void WriteLine(string message, string category) { Write(message + "\n", category); }
        public override void Write(string message) { Write(message, null); }
        public override void WriteLine(string message) { Write(message + "\n"); }
    }
}

Answer 1

在我的实验中，结构控制器似乎在角色缩放操作期间（角色完全删除之前）从DB服务器的动态白名单中删除角色IP。

如果这也是导致问题的原因，那么可能的解决方法是手动将IP范围0.0.0.0-255.255.255.255添加到数据库服务器的白名单中（以牺牲一些安全性为代价）。或者，您可以重新设计应用程序，以便在OnStop期间将数据/消息写入队列而不是DB（以便稍后将工作者角色复制到DB）。