|
| 1 | +// |
| 2 | +// Copyright (c) Microsoft Corporation. All rights reserved. |
| 3 | +// SPDX-License-Identifier: Apache-2.0 |
| 4 | +// |
| 5 | +// TEMPORARY CI DIAGNOSTIC -- not for merge. |
| 6 | +// Dumps native stacks of every thread when a single test exceeds a deadline, |
| 7 | +// so a CI-only hang can be root-caused without a debugger on the runner. |
| 8 | +// |
| 9 | +#ifdef _WIN32 |
| 10 | + |
| 11 | +#include <windows.h> |
| 12 | +#include <dbghelp.h> |
| 13 | +#include <tlhelp32.h> |
| 14 | + |
| 15 | +#include <atomic> |
| 16 | +#include <chrono> |
| 17 | +#include <cstdio> |
| 18 | +#include <cstdlib> |
| 19 | +#include <thread> |
| 20 | +#include <vector> |
| 21 | + |
| 22 | +#pragma comment(lib, "dbghelp.lib") |
| 23 | + |
| 24 | +namespace { |
| 25 | + |
| 26 | +std::atomic<long long> g_testStartMs{0}; |
| 27 | + |
| 28 | +long long NowMs() |
| 29 | +{ |
| 30 | + return std::chrono::duration_cast<std::chrono::milliseconds>( |
| 31 | + std::chrono::steady_clock::now().time_since_epoch()) |
| 32 | + .count(); |
| 33 | +} |
| 34 | + |
| 35 | +void DumpAllThreadStacks() |
| 36 | +{ |
| 37 | + HANDLE proc = ::GetCurrentProcess(); |
| 38 | + ::SymSetOptions(SYMOPT_UNDNAME | SYMOPT_LOAD_LINES); |
| 39 | + char exeDir[MAX_PATH] = {0}; |
| 40 | + ::GetModuleFileNameA(NULL, exeDir, MAX_PATH); |
| 41 | + char* lastSlash = strrchr(exeDir, '\\'); |
| 42 | + if (lastSlash != nullptr) |
| 43 | + { |
| 44 | + *lastSlash = '\0'; |
| 45 | + } |
| 46 | + ::SymInitialize(proc, exeDir, TRUE); |
| 47 | + |
| 48 | + DWORD pid = ::GetCurrentProcessId(); |
| 49 | + DWORD selfTid = ::GetCurrentThreadId(); |
| 50 | + HANDLE snap = ::CreateToolhelp32Snapshot(TH32CS_SNAPTHREAD, 0); |
| 51 | + if (snap == INVALID_HANDLE_VALUE) |
| 52 | + { |
| 53 | + printf("watchdog: CreateToolhelp32Snapshot failed %lu\n", ::GetLastError()); |
| 54 | + return; |
| 55 | + } |
| 56 | + |
| 57 | + THREADENTRY32 te; |
| 58 | + te.dwSize = sizeof(te); |
| 59 | + if (!::Thread32First(snap, &te)) |
| 60 | + { |
| 61 | + ::CloseHandle(snap); |
| 62 | + return; |
| 63 | + } |
| 64 | + |
| 65 | + do |
| 66 | + { |
| 67 | + if (te.th32OwnerProcessID != pid || te.th32ThreadID == selfTid) |
| 68 | + { |
| 69 | + continue; |
| 70 | + } |
| 71 | + |
| 72 | + HANDLE th = ::OpenThread( |
| 73 | + THREAD_GET_CONTEXT | THREAD_QUERY_INFORMATION | THREAD_SUSPEND_RESUME, |
| 74 | + FALSE, te.th32ThreadID); |
| 75 | + if (th == NULL) |
| 76 | + { |
| 77 | + continue; |
| 78 | + } |
| 79 | + |
| 80 | + std::vector<DWORD64> frames; |
| 81 | + ::SuspendThread(th); |
| 82 | + CONTEXT ctx; |
| 83 | + memset(&ctx, 0, sizeof(ctx)); |
| 84 | + ctx.ContextFlags = CONTEXT_FULL; |
| 85 | + if (::GetThreadContext(th, &ctx)) |
| 86 | + { |
| 87 | + STACKFRAME64 f; |
| 88 | + memset(&f, 0, sizeof(f)); |
| 89 | +#ifdef _M_IX86 |
| 90 | + DWORD machine = IMAGE_FILE_MACHINE_I386; |
| 91 | + f.AddrPC.Offset = ctx.Eip; |
| 92 | + f.AddrFrame.Offset = ctx.Ebp; |
| 93 | + f.AddrStack.Offset = ctx.Esp; |
| 94 | +#else |
| 95 | + DWORD machine = IMAGE_FILE_MACHINE_AMD64; |
| 96 | + f.AddrPC.Offset = ctx.Rip; |
| 97 | + f.AddrFrame.Offset = ctx.Rbp; |
| 98 | + f.AddrStack.Offset = ctx.Rsp; |
| 99 | +#endif |
| 100 | + f.AddrPC.Mode = AddrModeFlat; |
| 101 | + f.AddrFrame.Mode = AddrModeFlat; |
| 102 | + f.AddrStack.Mode = AddrModeFlat; |
| 103 | + |
| 104 | + for (int i = 0; i < 64; i++) |
| 105 | + { |
| 106 | + if (!::StackWalk64(machine, proc, th, &f, &ctx, NULL, |
| 107 | + ::SymFunctionTableAccess64, ::SymGetModuleBase64, NULL)) |
| 108 | + { |
| 109 | + break; |
| 110 | + } |
| 111 | + if (f.AddrPC.Offset == 0) |
| 112 | + { |
| 113 | + break; |
| 114 | + } |
| 115 | + frames.push_back(f.AddrPC.Offset); |
| 116 | + } |
| 117 | + } |
| 118 | + ::ResumeThread(th); |
| 119 | + ::CloseHandle(th); |
| 120 | + |
| 121 | + printf("\n--- thread %lu (%zu frames) ---\n", |
| 122 | + static_cast<unsigned long>(te.th32ThreadID), frames.size()); |
| 123 | + for (size_t i = 0; i < frames.size(); i++) |
| 124 | + { |
| 125 | + alignas(8) char buf[sizeof(SYMBOL_INFO) + 512]; |
| 126 | + memset(buf, 0, sizeof(buf)); |
| 127 | + SYMBOL_INFO* sym = reinterpret_cast<SYMBOL_INFO*>(buf); |
| 128 | + sym->SizeOfStruct = sizeof(SYMBOL_INFO); |
| 129 | + sym->MaxNameLen = 500; |
| 130 | + |
| 131 | + char modName[MAX_PATH] = "?"; |
| 132 | + DWORD64 modBase = 0; |
| 133 | + IMAGEHLP_MODULE64 mi; |
| 134 | + memset(&mi, 0, sizeof(mi)); |
| 135 | + mi.SizeOfStruct = sizeof(mi); |
| 136 | + int symType = -1; |
| 137 | + if (::SymGetModuleInfo64(proc, frames[i], &mi)) |
| 138 | + { |
| 139 | + strncpy_s(modName, mi.ModuleName, _TRUNCATE); |
| 140 | + modBase = mi.BaseOfImage; |
| 141 | + symType = static_cast<int>(mi.SymType); |
| 142 | + } |
| 143 | + unsigned long long rva = (modBase != 0) |
| 144 | + ? static_cast<unsigned long long>(frames[i] - modBase) : 0ull; |
| 145 | + |
| 146 | + DWORD64 disp = 0; |
| 147 | + if (::SymFromAddr(proc, frames[i], &disp, sym)) |
| 148 | + { |
| 149 | + printf(" %02zu %s+0x%llx %s!%s+0x%llx\n", i, modName, rva, modName, sym->Name, |
| 150 | + static_cast<unsigned long long>(disp)); |
| 151 | + } |
| 152 | + else |
| 153 | + { |
| 154 | + printf(" %02zu %s+0x%llx (symtype=%d)\n", i, modName, rva, symType); |
| 155 | + } |
| 156 | + } |
| 157 | + fflush(stdout); |
| 158 | + } while (::Thread32Next(snap, &te)); |
| 159 | + |
| 160 | + ::CloseHandle(snap); |
| 161 | +} |
| 162 | + |
| 163 | +} // namespace |
| 164 | + |
| 165 | +void FuncTestsWatchdog_NoteTestStart() |
| 166 | +{ |
| 167 | + g_testStartMs.store(NowMs()); |
| 168 | +} |
| 169 | + |
| 170 | +void FuncTestsWatchdog_Start() |
| 171 | +{ |
| 172 | + const char* env = nullptr; |
| 173 | + size_t len = 0; |
| 174 | + char buffer[32] = {0}; |
| 175 | + if (::getenv_s(&len, buffer, sizeof(buffer), "FUNCTESTS_WATCHDOG_SEC") == 0 && len > 1) |
| 176 | + { |
| 177 | + env = buffer; |
| 178 | + } |
| 179 | + unsigned seconds = (env != nullptr) ? static_cast<unsigned>(atoi(env)) : 0u; |
| 180 | + if (seconds == 0) |
| 181 | + { |
| 182 | + return; |
| 183 | + } |
| 184 | + |
| 185 | + g_testStartMs.store(NowMs()); |
| 186 | + std::thread([seconds]() { |
| 187 | + for (;;) |
| 188 | + { |
| 189 | + std::this_thread::sleep_for(std::chrono::seconds(5)); |
| 190 | + long long started = g_testStartMs.load(); |
| 191 | + if (started == 0) |
| 192 | + { |
| 193 | + continue; |
| 194 | + } |
| 195 | + if ((NowMs() - started) > static_cast<long long>(seconds) * 1000) |
| 196 | + { |
| 197 | + printf("\n#### WATCHDOG: current test exceeded %u s -- dump #1 ####\n", seconds); |
| 198 | + fflush(stdout); |
| 199 | + DumpAllThreadStacks(); |
| 200 | + std::this_thread::sleep_for(std::chrono::seconds(45)); |
| 201 | + printf("\n#### WATCHDOG: dump #2 (45 s later; identical stacks == truly stuck) ####\n"); |
| 202 | + fflush(stdout); |
| 203 | + DumpAllThreadStacks(); |
| 204 | + printf("\n#### WATCHDOG: terminating process ####\n"); |
| 205 | + fflush(stdout); |
| 206 | + ::TerminateProcess(::GetCurrentProcess(), 99); |
| 207 | + } |
| 208 | + } |
| 209 | + }).detach(); |
| 210 | +} |
| 211 | + |
| 212 | +#else |
| 213 | +void FuncTestsWatchdog_NoteTestStart() {} |
| 214 | +void FuncTestsWatchdog_Start() {} |
| 215 | +#endif |
0 commit comments