C# 解決死結

處理這個死結問題，花了好幾天，相信遇到的同學，一樣頭疼，但有個好輔助類的話（好在.net的API足夠強大），就沒這麼頭疼了 注意

本篇文章的解決方案只適合使用lock(obj)，或是：Monitor.Enter(obj); …. Monitor.Exit(obj)的方式
類似醬紫的死結

如果你使用的是：AutoResetEvent.Set/Rest, Monitor.Wait/Pulse, Mutex的方式，請另尋他法。 輔助類

//------------------------------------------------------------------------// Craeted by Jave.Lin 4/21/2018 5:31:57 PM//------------------------------------------------------------------------using System;using System.Collections.Generic;using System.Threading;//namespace Common.ComUtil//{    /// <summary>    /// Locker Information    /// author  :   Jave.Lin    /// date    :   4/21/2018 5:31:57 PM    /// </summary>    public class Locker    {        public object tag; // 附帶上下文資料        public int threadID; // 擷取鎖的線程ID        public string name; // 鎖的名稱        public int lockedTimes; // 累積擷取鎖多少次，便於分析死結幾率        public int lockingTs; // 擷取鎖那刻的時間戳記        public bool enter; // 擷取鎖的標記        public bool exit; // 釋放鎖的標記        public string lockingStackTrace; // 擷取鎖那刻的調用方法棧，下文Dumps資訊執行個體可以看到很清楚        public int lockingDLTs; // 識別為死結時設定的時間戳記        public List<Locker> pRecordList;        public Locker(string name)        {            this.name = name;            pRecordList = new List<Locker>();        }        public void PushRecord()        {            pRecordList.Add(RetrieveRecord());        }        public void ClearRecord()        {            pRecordList.Clear();        }        public Locker RetrieveRecord()        {            var ret = new Locker(this.name);            ret.tag = tag;            ret.threadID = threadID;            ret.name = name;            ret.lockedTimes = lockedTimes;            ret.lockingTs = lockingTs;            ret.enter = enter;            ret.exit = exit;            ret.lockingStackTrace = lockingStackTrace;            ret.lockingDLTs = lockingDLTs;            return ret;        }        public override string ToString()        {            string[] strs = new string[]            {                "ThreadId:" + threadID,                "Name:" + name,                "LockedTimes:" + lockedTimes,                "LockingTs:" + lockingTs,                "LocingEt:" + (lockingDLTs - lockingTs) + ("(ms"),                "Enter:" + enter,                "Exit:" + exit,                "Tag:" + (tag != null ? tag.ToString() : "null"),                "LockingStackTrace:\n" + lockingStackTrace,                (pRecordList.Count > 0 ? "LockedRecord:\n\t" + string.Join("\t-record--------------", pRecordList) : "")            };            return string.Join("\r\n", strs);        }    }    public delegate void OnDL();    /// <summary>    /// Check Dead Lock (CDL)    /// author  :   Jave.Lin    /// date    :   4/21/2018 5:31:57 PM    /// </summary>    public static class CDL    {        // had been locked in map        public static readonly Dictionary<Locker, bool> _s_pLockedMap = new Dictionary<Locker, bool>();        // the last         public static readonly Dictionary<Locker, List<Locker>> _s_pLockingMap = new Dictionary<Locker, List<Locker>>();        public const bool THROW_ER = true;        public const int DEAD_LOCK_TIME_OUT = 3000; // 這個閥值按需調整，實際的線上產品伺服器程式如果負載過大時，可能也會有部分任務處理過久，導致‘取鎖’等待過久    public static event OnDL OnDLEvent;        private static void _PushToWaitQueue(Locker locker)        {            List<Locker> list = null;            if (!_s_pLockingMap.TryGetValue(locker, out list))            {                list = new List<Locker>();                _s_pLockingMap[locker] = list;            }            list.Add(locker.RetrieveRecord());        }        private static void _ClearFromLocking(Locker locker)        {            List<Locker> list = null;            if (_s_pLockingMap.TryGetValue(locker, out list))            {                list.Clear();                _s_pLockingMap.Remove(locker);            }        }        private static void _BeforeEnter(Locker locker)        {            if (locker.enter)            {                locker.PushRecord();                _PushToWaitQueue(locker);            }        }        private static void _Enter(Locker locker)        {            locker.enter = true;            locker.exit = false;            locker.lockingTs = Environment.TickCount;            locker.threadID = Thread.CurrentThread.ManagedThreadId;            locker.lockingStackTrace = GetCurStackTrace("->\n");            Interlocked.Increment(ref locker.lockedTimes);            _s_pLockedMap[locker] = true;        }        private static void _Exit(Locker locker)        {            if (!Monitor.IsEntered(locker))            {                locker.lockingTs = Environment.TickCount;                var msg = "!Monitor.IsEntered(locker)";                if (THROW_ER)                {                    throw new Exception(msg);                }                else                {                    _WarningWriteLine(msg);                }            }            else            {                locker.exit = true;                _s_pLockedMap.Remove(locker);                _ClearFromLocking(locker);                locker.ClearRecord();                Monitor.Exit(locker);            }        }        private static string _GetWaitQueue(Locker locker)        {            if (_s_pLockingMap.ContainsKey(locker))            {                return string.Join("\n@@@@@", _s_pLockingMap[locker]);            }            return "";        }        public static string Dumps()        {            var itemList = new List<Locker>();            var contentList = new List<string>();            foreach (var item in _s_pLockedMap)            {                itemList.Add(item.Key);            }            itemList.Sort((a, b) =>            {                return (b.lockingDLTs - b.lockingTs) - (a.lockingDLTs - a.lockingTs);            });            foreach (var item in itemList)            {                contentList.Add(item.ToString() + "\n$$$$$$$$$$Before Locking WaitQueue$$$$$$$$\n" + _GetWaitQueue(item));            }            return string.Join("\r\n=line============\r\n", contentList);        }        public static void CheckDL(Locker locker, Action actoin)        {            try            {                _BeforeEnter(locker);                if (Monitor.TryEnter(locker, DEAD_LOCK_TIME_OUT))                {                    _Enter(locker);                    actoin.Invoke();                }                else                {                    locker.lockingDLTs = Environment.TickCount;                    _WarningWriteLine("TryEnter time out");                    if (THROW_ER)                    {                        _ShowGetLockTimeout();                    }                    else                    {                        actoin.Invoke();                    }                }            }            catch (Exception e)            {                _ErrorWriteLine(e.ToString());            }            finally            {                _Exit(locker);            }        }        public static T CheckDL<T>(Locker locker, Func<T> actoin)        {            T ret = default(T);            try            {                _BeforeEnter(locker);                if (Monitor.TryEnter(locker, DEAD_LOCK_TIME_OUT))                {                    _Enter(locker);                    ret = actoin.Invoke();                }                else                {                    locker.lockingDLTs = Environment.TickCount;                    _WarningWriteLine("TryEnter time out");                    if (THROW_ER)                    {                        _ShowGetLockTimeout();                    }                    else                    {                        actoin.Invoke();                    }                }            }            catch (Exception e)            {                _ErrorWriteLine(e.ToString());            }            finally            {                _Exit(locker);            }            return ret;        }        public static string GetCurStackTrace(string separactor = "->")        {            System.Diagnostics.StackTrace st = new System.Diagnostics.StackTrace();            System.Diagnostics.StackFrame[] sfs = st.GetFrames();            List<string> methodNameList = new List<string>();            for (int i = 1; i < sfs.Length; ++i)            {                if (System.Diagnostics.StackFrame.OFFSET_UNKNOWN == sfs[i].GetILOffset()) break;                var m = sfs[i].GetMethod();                var dn = m.DeclaringType.Name;                var mn = m.Name;                methodNameList.Add(new string(' ', sfs.Length - i) + dn + "::" + mn + "()");            }            methodNameList.Reverse();            return string.Join(separactor, methodNameList);        }        private static void _ShowGetLockTimeout()        {            OnDLEvent?.Invoke();            var msg = "!!!!!!!!!!!!!!!!!!DeadLock!!!!!!!!!!!!!!!!!!!!!!!!!!";            _ErrorWriteLine(msg);            throw new Exception(msg);        }        private static void _ErrorWriteLine(string msg, params object[] args)        {            var srcColor = Console.ForegroundColor;            Console.ForegroundColor = ConsoleColor.Red;            Console.WriteLine(msg, args);            Console.ForegroundColor = srcColor;        }        private static void _WarningWriteLine(string msg, params object[] args)        {            var srcColor = Console.ForegroundColor;            Console.ForegroundColor = ConsoleColor.Yellow;            Console.WriteLine(msg, args);            Console.ForegroundColor = srcColor;        }    }//}

使用方法要點、注意副作用、處理副作用 使用方法要點

// 我們平常用的：lock(obj){ // code here}改寫成：CDL.CheckDL(obj, ()=>{ // code here});// 怎麼方便改，是個問題// 用到的lock少的話，手動一個個改吧// 如果巨量的話，建議CTRL + SHIFT + H來批量替換Lock的代碼吧（寫個正則）// 將：CDL的namespace去掉，這樣就不用導namespace了。

一旦有死結出現，那麼將會命中CDL::_ShowGetLockTimeout方法
然後將CDL.Dumps()的內容列印出來，就可以知道，當前哪些CDL.CheckDL的地方有死結。
Dumps很詳細，具體還可以根據自己的需要來對 Locker的資訊做調整。 從dumps資訊中辨別死結

Dumps資訊中，辨別哪些是死結狀態的，看：LockingEt的值是多少就知道了
LockingEt是：Locking Elapsed Time的意思，擷取鎖多長時間了
LockingEt值只要大於零，且接近於：CDL.DEAD_LOCK_TIME_OUT的值，都基本上是死結 注意副作用

明顯原來的代碼邏輯會發生改變
多了一些方法調用
特別是將原來的代碼的位置，改變了，放到了一個lambda（其實在IL中是個匿名函數）
先不說結構上變化了，效率也會有丟丟影響的。 處理副作用

還有一個方法可以去出副作用，就是再寫個工具，將編譯出來的DLL，批量處理CDL.CheckDL的IL代碼，改為原來的lock(obj)方式，當然，前提是先需要大量測試後再用這個工具處理，不然如果中途還是出現了死結的話，定位問題還是會很頭疼的。（製作這個工具，理論上是可以的，但需要對IL熟悉） 注意CDL.DEAD_LOCK_TIME_OUT

CDL.DEAD_LOCK_TIME_OUT= 3000; // 這個閥值按需調整，實際的線上產品伺服器程式如果負載過大時，可能也會有部分任務處理過久，導致‘取鎖’等待過久，所以出現Monitor.TryEnter timeout時，不一定是死結。
你可以按你的需求來調整該值，如：調整個：60000（60秒），意思是你確定了，取鎖時間逾時為60秒的，都算是有死結任務導致 CDL.Dumps內容執行個體

ThreadId:10Name:CBCServerAliveLockerLockedTimes:1LockingTs:7496815LockingEt:-7496815(ms // =========辨別死結=======負數的都不用看Enter:TrueExit:FalseTag:nullLockingStackTrace:         Program::Main()->          XXXServer::WaitForExit()->           CDL::CheckDL()=line============ThreadId:15Name:CTcpNetworkerLockedTimes:486LockingTs:7519248LockingEt:2995(ms // =辨別死結=此乃死結也，接近CDL.DEAD_LOCK_TIME_OUT（3000 MS）的值，由於CBattleRoom的Locker線程ID為6的鎖逾時而導致的Enter:TrueExit:FalseTag:nullLockingStackTrace: _IOCompletionCallback::PerformIOCompletionCallback()->  BaseOverlappedAsyncResult::CompletionPortCallback()->   LazyAsyncResult::ProtectedInvokeCallback()->    ContextAwareResult::Complete()->     ExecutionContext::Run()->      ExecutionContext::Run()->       ExecutionContext::RunInternal()->        ContextAwareResult::CompleteCallback()->         LazyAsyncResult::Complete()->          XXXNetworker::_OnBeginReceiveCallback()->           CDL::CheckDL()=line============ThreadId:15Name:CEventMgrLockedTimes:247LockingTs:7519248LockingEt:-7519248(ms // =========辨別死結=======負數的都不用看Enter:TrueExit:FalseTag:nullLockingStackTrace: _IOCompletionCallback::PerformIOCompletionCallback()->  BaseOverlappedAsyncResult::CompletionPortCallback()->   LazyAsyncResult::ProtectedInvokeCallback()->    ContextAwareResult::Complete()->     ExecutionContext::Run()->      ExecutionContext::Run()->       ExecutionContext::RunInternal()->        ContextAwareResult::CompleteCallback()->         LazyAsyncResult::Complete()->          XXXNetworker::_OnBeginReceiveCallback()->           CDL::CheckDL()->            <>c__DisplayClass78_0::<_OnBeginReceiveCallback>b__0()->             XXXConnection::XXXNetworker_OnPackageEvent()->              CEventMgr::Invoke()->               CDL::CheckDL()=line============ThreadId:6Name:CBattleRoomMgrLockedTimes:689LockingTs:7519248LockingEt:-7519248(ms // =========辨別死結=======負數的都不用看Enter:TrueExit:FalseTag:nullLockingStackTrace: _ThreadPoolWaitCallback::PerformWaitCallback()->  ThreadPoolWorkQueue::Dispatch()->   Task::System.Threading.IThreadPoolWorkItem.ExecuteWorkItem()->    Task::ExecuteEntry()->     Task::ExecuteWithThreadLocal()->      ExecutionContext::Run()->       ExecutionContext::RunInternal()->        Task::ExecutionContextCallback()->         Task::Execute()->          Task::InnerInvoke()->           XXXServer::<StartUp>b__36_0()->            CDL::CheckDL()->             XXXServer::<StartUp>b__36_1()->              XXXServer::_DeadLockMethod()->               XXXServerInst::_DeadLockMethod1()->                CDL::CheckDL()=line============ThreadId:6Name:CBattleRoomLockedTimes:802LockingTs:7519248LockingEt:2995(ms // ==============辨別死結======此乃死結也，接近CDL.DEAD_LOCK_TIME_OUT（3000 MS）的值Enter:TrueExit:FalseTag:BroadcastAll3 startingLockingStackTrace: _ThreadPoolWaitCallback::PerformWaitCallback()->  ThreadPoolWorkQueue::Dispatch()->   Task::System.Threading.IThreadPoolWorkItem.ExecuteWorkItem()->    Task::ExecuteEntry()->     Task::ExecuteWithThreadLocal