.NET某消防物聯網後臺服務內存泄漏分析
一:背景
1. 講故事
去年十月份有位朋友從微信找到我,說他的程序內存要炸掉瞭。。。截圖如下:
時間有點久,圖片都被清理瞭,不過有點諷刺的是,自己的程序本身就是做監控的,結果自己出瞭問題,太尷尬瞭🤣🤣🤣
二:Windbg 分析
1. 托管還是非托管
這個是甄別內存問題的第一步,通過 !address -summary 和 !eeheap -gc 兩個命令基本就可以斷定。
0:000> !address -summary Mapping file section regions... Mapping module regions... Mapping PEB regions... Mapping TEB and stack regions... Mapping heap regions... Mapping page heap regions... Mapping other regions... Mapping stack trace database regions... Mapping activation context regions... --- Usage Summary ---------------- RgnCount ----------- Total Size -------- %ofBusy %ofTotal Free 237 7ffc`1e222000 ( 127.985 TB) 99.99% <unknown> 594 3`b9b20000 ( 14.901 GB) 95.96% 0.01% Heap 370 0`12a2a000 ( 298.164 MB) 1.88% 0.00% Image 1248 0`0ee5a000 ( 238.352 MB) 1.50% 0.00% Stack 315 0`06780000 ( 103.500 MB) 0.65% 0.00% Other 13 0`001d7000 ( 1.840 MB) 0.01% 0.00% TEB 105 0`000d2000 ( 840.000 kB) 0.01% 0.00% PEB 1 0`00001000 ( 4.000 kB) 0.00% 0.00% --- Type Summary (for busy) ------ RgnCount ----------- Total Size -------- %ofBusy %ofTotal MEM_PRIVATE 1178 3`ce03d000 ( 15.219 GB) 98.00% 0.01% MEM_IMAGE 1409 0`0f6fd000 ( 246.988 MB) 1.55% 0.00% MEM_MAPPED 59 0`04694000 ( 70.578 MB) 0.44% 0.00% --- State Summary ---------------- RgnCount ----------- Total Size -------- %ofBusy %ofTotal MEM_FREE 237 7ffc`1e222000 ( 127.985 TB) 99.99% MEM_COMMIT 2326 3`c7543000 ( 15.115 GB) 97.33% 0.01% MEM_RESERVE 320 0`1a88b000 ( 424.543 MB) 2.67% 0.00% 0:000> !eeheap -gc Number of GC Heaps: 1 generation 0 starts at 0x0000009902B57670 generation 1 starts at 0x0000009902A56810 generation 2 starts at 0x00000095318C1000 ephemeral segment allocation context: (0x0000009902D724A8, 0x0000009902D724C0) segment begin allocated committed allocated size committed size ... 00000098FFBE0000 00000098FFBE1000 0000009902D724A8 0000009902D7D000 0x31914a8(51975336) 0x319c000(52019200) Large object heap starts at 0x00000095418C1000 segment begin allocated committed allocated size committed size 00000095418C0000 00000095418C1000 00000095475D8D98 00000095475D9000 0x5d17d98(97615256) 0x5d18000(97615872) Total Allocated Size: Size: 0x398e6cbc8 (15450164168) bytes. Total Committed Size: Size: 0x398e7b000 (15450222592) bytes. ------------------------------ GC Allocated Heap Size: Size: 0x398e6cbc8 (15450164168) bytes. GC Committed Heap Size: Size: 0x398e7b000 (15450222592) bytes.
從輸出信息看,好傢夥。。。進程提交內存是 15G, 托管堆差不多也是 15G,這就說明當前是相對簡單的 托管內存泄漏。
2. 到底是什麼在泄漏
要想知道到底是什麼在泄漏,可以先到托管堆上看看有沒有什麼異常的對象,使用 !dumpheap -stat 命令。
0:000> !dumpheap -stat Statistics: MT Count TotalSize Class Name ... 00007ff8815d0f88 7260233 290409320 System.Collections.ArrayList 00007ff8815e6830 7313696 326240826 System.String 000000952fbbd2b0 12141398 509369998 Free 00007ff880685cf0 7254983 928637824 System.Diagnostics.ProcessInfo 00007ff88065f7d0 7256845 2031916600 System.Diagnostics.Process 00007ff8815e6ea8 7391338 2230744600 System.Object[] 00007ff88068fa70 186800748 8966435904 System.Diagnostics.ThreadInfo
從卦象上來看,真的太奇怪瞭,如果大傢瞭解 Process 類,應該知道 ProcessInfo 和 ThreadInfo 都是掛在 Process 下的,而且 ThreadInfo 對象高達 1.8億,真的太🐂👃瞭,看樣子程序是在不斷的做 Process.Start 操作吧。
接下來要探究的問題是 ThreadInfo 到底正在被誰持有??? 可以挑幾個看看它們的 !gcroot, 首尾法查瞭幾個,都是沒有引用根,如下所示:
0:000> !gcroot 0000009531e8f760 Found 0 unique roots (run '!GCRoot -all' to see all roots). 0:000> !gcroot 0000009531e8f670 Found 0 unique roots (run '!GCRoot -all' to see all roots). 0:000> !gcroot 0000009531e8f378 Found 0 unique roots (run '!GCRoot -all' to see all roots).
3. 無引用根為什麼不被回收
既然對象沒有引用根,為什麼 GC 不回收它呢? 這裡就需要談經驗瞭,在我之前分析的很多關於內存泄漏 的dump中,我都是從 生產端 找問題,貌似還沒有出現一個從 消費端 找問題的案例,參考模型如下:
生產者既然沒問題,那消費端能有什麼問題呢?大傢可以想一想,托管堆的消費端起碼有兩個角色。
- GC
- Finalizer 線程
GC 肯定是不會出問題的,那就隻能懷疑 Finalizer 線程出瞭什麼問題,可以用 !t 命令把所有線程調出來看看。
0:000> !t ThreadCount: 9566 UnstartedThread: 0 BackgroundThread: 88 PendingThread: 0 DeadThread: 9471 Hosted Runtime: no Lock ID OSID ThreadOBJ State GC Mode GC Alloc Context Domain Count Apt Exception 0 1 40e18 000000952fbdfe50 26020 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 STA 2 2 41ce8 000000952fc0e4d0 2b220 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 MTA (Finalizer) 4 4 41cb4 000000954c324970 27220 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 STA 5 5 41cb8 000000954c36d1e0 2027220 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 STA 6 6 41c58 000000954c33f070 2027220 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 STA 7 7 41c38 000000954c33f840 27220 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 STA 8 8 41e0c 000000954c333580 27220 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 STA 9 9 41e2c 000000954c354440 27220 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 STA 10 10 41f24 000000954c355840 27220 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 STA ... XXXX 9446 0 000000957a233410 8039820 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 MTA (Threadpool Completion Port) XXXX 9447 0 0000009579f83e30 8039820 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 MTA (Threadpool Completion Port) XXXX 9450 0 000000957a46dcf0 8039820 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 MTA (Threadpool Completion Port) XXXX 9449 0 000000957967c6e0 8039820 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 MTA (Threadpool Completion Port) XXXX 9448 0 000000957aee0010 8039820 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 MTA (Threadpool Completion Port) XXXX 9452 0 00000095796824a0 8039820 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 MTA (Threadpool Completion Port) XXXX 9451 0 000000957af05df0 8039820 Preemptive 0000000000000000:0000000000000000 000000952fbd2e40 0 MTA (Threadpool Completion Port)
這不看不知道,一看嚇一跳,當前進程有 9566 個,死線程高達 9471 個 ,以過往經驗,這小子不用線程池,用 new Thread 咯, 🐂👃🦆 ,吐槽結束,再看下 Finalizer 線程正在做什麼,使用 ~2s & !dumpstack
0:002> ~2s ntdll!NtWaitForSingleObject+0xa: 00007ff8`8e220c8a c3 ret 0:002> !dumpstack OS Thread Id: 0x41ce8 (2) Current frame: ntdll!NtWaitForSingleObject+0xa Child-SP RetAddr Caller, Callee 0000009549e4e120 00007ff88b591118 KERNELBASE!WaitForSingleObjectEx+0x94, calling ntdll!NtWaitForSingleObject 0000009549e4e1c0 00007ff88da3e334 combase!MTAThreadWaitForCall+0x54 [d:\9147\com\combase\dcomrem\channelb.cxx:5657], calling KERNELBASE!WaitForSingleObject 0000009549e4e210 00007ff88d8fe089 combase!MTAThreadDispatchCrossApartmentCall+0x75 [d:\9147\com\combase\dcomrem\chancont.cxx:193], calling combase!MTAThreadWaitForCall [d:\9147\com\combase\dcomrem\channelb.cxx:5619] 0000009549e4e240 00007ff88da3e13d combase!CRpcChannelBuffer::SendReceive2+0x64b [d:\9147\com\combase\dcomrem\channelb.cxx:4796], calling combase!MTAThreadDispatchCrossApartmentCall [d:\9147\com\combase\dcomrem\chancont.cxx:156] 0000009549e4e2b0 00007ff88e1bb6f7 ntdll!RtlAllocateHeap+0xd7, calling ntdll!RtlpLowFragHeapAllocFromContext ... 0000009549e4f5d0 00007ff8827d79cd clr!ManagedThreadBase_DispatchOuter+0x75, calling clr!ManagedThreadBase_DispatchMiddle 0000009549e4f5e0 00007ff8828601af clr!EEConfig::GetConfigDWORD_DontUse_+0x3b, calling clr!EEConfig::GetConfiguration_DontUse_ 0000009549e4f660 00007ff8828574fa clr!FinalizerThread::FinalizerThreadStart+0x10a, calling clr!ManagedThreadBase_DispatchOuter 0000009549e4f6a0 00007ff8827d55b9 clr!EEHeapFreeInProcessHeap+0x45, calling kernel32!HeapFreeStub 0000009549e4f700 00007ff882882e8f clr!Thread::intermediateThreadProc+0x86 0000009549e4f780 00007ff882882e6f clr!Thread::intermediateThreadProc+0x66, calling clr!_chkstk 0000009549e4f7c0 00007ff88dcd13d2 kernel32!BaseThreadInitThunk+0x22 0000009549e4f7f0 00007ff88e2003c4 ntdll!RtlUserThreadStart+0x34
從堆棧信息看,原來是終結器線程卡死瞭,從 MTAThreadDispatchCrossApartmentCall方法看,貌似是 MTA 向 STA 做瞭一個調用,到這裡有經驗的朋友應該知道,這和 com 組件有關系瞭,也就是說 Finalizer 線程無法釋放由 STA 線程創建的 COM 對象,那到底是哪一個線程創建瞭沒有被合理釋放的 COM 組件呢?
4. 尋找創建 COM 組件的線程
說實話,這個對COM組件不瞭解的話,很難找出答案,但天無絕人之路,當我回頭看 線程列表 的時候,發現居然有 38 個 STA線程,截圖如下:
這裡面肯定有問題,接下來抽一個線程看看調用棧如何。
0:004> !clrstack OS Thread Id: 0x41cb4 (4) Child SP IP Call Site 000000954da1ee38 00007ff88e220c8a [HelperMethodFrame: 000000954da1ee38] System.Threading.Thread.SleepInternal(Int32) 000000954da1ef30 00007ff88138c20a *** WARNING: Unable to verify checksum for mscorlib.ni.dll System.Threading.Thread.Sleep(Int32) 000000954da1ef60 00007ff82322437f xxx.CFileLogTask.DoWork() 000000954da1f160 00007ff8232234d6 xxx.CTask.InitStart() 000000954da1f240 00007ff8813c31d3 System.Threading.ExecutionContext.RunInternal(System.Threading.ExecutionContext, System.Threading.ContextCallback, System.Object, Boolean) 000000954da1f310 00007ff8813c3064 System.Threading.ExecutionContext.Run(System.Threading.ExecutionContext, System.Threading.ContextCallback, System.Object, Boolean) 000000954da1f340 00007ff8813c3032 System.Threading.ExecutionContext.Run(System.Threading.ExecutionContext, System.Threading.ContextCallback, System.Object) 000000954da1f390 00007ff8813bc812 System.Threading.ThreadHelper.ThreadStart() 000000954da1f5e8 00007ff8827d6bb3 [GCFrame: 000000954da1f5e8] 000000954da1f938 00007ff8827d6bb3 [DebuggerU2MCatchHandlerFrame: 000000954da1f938]
接下來反編譯下 xxx.CFileLogTask.DoWork() 方法看看它是如何被 Thread 執行的。
到這裡終於水落石出,罪魁禍首在 CurrThread.SetApartmentState(ApartmentState.STA); 這一句上,我也不知道為啥開個 Thread 還要給個 STA。。。
三:總結
本次事故主要是因為在 STA 線程上用到瞭 COM 組件,導致讓 MTA 模型的 Finalizer 線程去釋放時被卡死,而這個Thread又沒有用 Application.Run 啟動消息循環,STA也是 Sleep 狀態,我個人感覺兩者無法通訊,給到朋友的建議是去掉 Thread 的 STA。
其實這裡有一個很好的點就是,當內存暴漲,不一定是 生產端 的問題,也有可能是 消費端。
以上就是.NET某消防物聯網後臺服務內存泄漏分析的詳細內容,更多關於.NET後臺服務內存泄漏的資料請關註WalkonNet其它相關文章!
推薦閱讀:
- .NET Windbg分析某婦產醫院WPF內存溢出
- C#非托管泄漏中HEAP_ENTRY的Size對不上解析
- C#程序終極調試實現windbg的時間旅行
- .NET新能源汽車鋰電池檢測程序UI掛死問題分析
- .NET CPU爆高事故事故分析某供應鏈WEB網站