已合并
fix：修复dump文件未完整落盘 #3266
yring_8创建于 6月30日
fix：修复dump文件未完整落盘 #3266
已合并
从已删除 :9.1.0合入到cann/runtime9.1.0
共 15 个文件变更+356-134
| @@ -38,7 +38,10 @@ public: | |||
| 38 | bool Push(T &value) | 38 | bool Push(T &value) |
| 39 | { | 39 | { |
| 40 | std::unique_lock<std::mutex> lk(mtx_); | 40 | std::unique_lock<std::mutex> lk(mtx_); |
| 41 | - cvPop_.wait(lk, [this] { return !this->IsFull() || this->quit_;}); | 41 | + cvPop_.wait(lk, [this] { return !this->IsFullUnlocked() || this->quit_;}); |
| 42 | + if (this->quit_) { | ||
| 43 | + return false; | ||
| 44 | + } | ||
| 42 | dataQueue_.push(value); | 45 | dataQueue_.push(value); |
| 43 | cvPush_.notify_all(); | 46 | cvPush_.notify_all(); |
| 44 | return true; | 47 | return true; |
| @@ -47,8 +50,8 @@ public: | |||
| 47 | bool Pop(T &value) | 50 | bool Pop(T &value) |
| 48 | { | 51 | { |
| 49 | std::unique_lock<std::mutex> lk(mtx_); | 52 | std::unique_lock<std::mutex> lk(mtx_); |
| 50 | - cvPush_.wait(lk, [this] { return !this->IsEmpty() || this->quit_; }); | 53 | + cvPush_.wait(lk, [this] { return !this->IsEmptyUnlocked() || this->quit_; }); |
| 51 | - if (!this->IsEmpty()) { | 54 | + if (!this->IsEmptyUnlocked()) { |
| 52 | value = this->dataQueue_.front(); | 55 | value = this->dataQueue_.front(); |
| 53 | this->dataQueue_.pop(); | 56 | this->dataQueue_.pop(); |
| 54 | cvPop_.notify_all(); | 57 | cvPop_.notify_all(); |
| @@ -60,10 +63,49 @@ public: | |||
| 60 | 63 | ||
| 61 | bool IsEmpty() const | 64 | bool IsEmpty() const |
| 62 | { | 65 | { |
| 63 | - return dataQueue_.empty(); | 66 | + std::lock_guard<std::mutex> lk(mtx_); |
| 67 | + return IsEmptyUnlocked(); | ||
| 64 | } | 68 | } |
| 65 | 69 | ||
| 66 | bool IsFull() const | 70 | bool IsFull() const |
| 71 | + { | ||
| 72 | + std::lock_guard<std::mutex> lk(mtx_); | ||
| 73 | + return IsFullUnlocked(); | ||
| 74 | + } | ||
| 75 | + | ||
| 76 | + void Init() | ||
| 77 | + { | ||
| 78 | + std::lock_guard<std::mutex> lk(mtx_); | ||
| 79 | + quit_ = false; | ||
| 80 | + std::queue<T>().swap(dataQueue_); | ||
| 81 | + } | ||
| 82 | + | ||
| 83 | + void Quit() | ||
| 84 | + { | ||
| 85 | + std::lock_guard<std::mutex> lk(mtx_); | ||
| 86 | + quit_ = true; | ||
| 87 | + cvPush_.notify_all(); | ||
| 88 | + cvPop_.notify_all(); | ||
| 89 | + } | ||
| 90 | + | ||
| 91 | + uint32_t Size() const | ||
| 92 | + { | ||
| 93 | + std::lock_guard<std::mutex> lk(mtx_); | ||
| 94 | + return dataQueue_.size(); | ||
| 95 | + } | ||
| 96 | + | ||
| 97 | + void SetPath(std::string path) | ||
| 98 | + { | ||
| 99 | + path_ = path; | ||
| 100 | + } | ||
| 101 | + | ||
| 102 | +private: | ||
| 103 | + bool IsEmptyUnlocked() const | ||
| 104 | + { | ||
| 105 | + return dataQueue_.empty(); | ||
| 106 | + } | ||
| 107 | + | ||
| 108 | + bool IsFullUnlocked() const | ||
| 67 | { | 109 | { |
| 68 | struct sysinfo info; | 110 | struct sysinfo info; |
| 69 | const size_t queueSize = 60; | 111 | const size_t queueSize = 60; |
| @@ -84,33 +126,6 @@ public: | |||
| 84 | return (info.freeram < (info.totalram * (1 - ADX_QUEUE_FULL_SIZE))) && dataQueue_.size() > queueSize; | 126 | return (info.freeram < (info.totalram * (1 - ADX_QUEUE_FULL_SIZE))) && dataQueue_.size() > queueSize; |
| 85 | } | 127 | } |
| 86 | 128 | ||
| 87 | - void Init() | ||
| 88 | - { | ||
| 89 | - std::lock_guard<std::mutex> lk(mtx_); | ||
| 90 | - quit_ = false; | ||
| 91 | - } | ||
| 92 | - | ||
| 93 | - void Quit() | ||
| 94 | - { | ||
| 95 | - std::lock_guard<std::mutex> lk(mtx_); | ||
| 96 | - if (!quit_) { | ||
| 97 | - quit_ = true; | ||
| 98 | - cvPush_.notify_all(); | ||
| 99 | - cvPop_.notify_all(); | ||
| 100 | - } | ||
| 101 | - } | ||
| 102 | - | ||
| 103 | - uint32_t Size() const | ||
| 104 | - { | ||
| 105 | - return dataQueue_.size(); | ||
| 106 | - } | ||
| 107 | - | ||
| 108 | - void SetPath(std::string path) | ||
| 109 | - { | ||
| 110 | - path_ = path; | ||
| 111 | - } | ||
| 112 | - | ||
| 113 | -private: | ||
| 114 | uint64_t InitMemLimit() const { | 129 | uint64_t InitMemLimit() const { |
| 115 | std::ifstream memLimitV1(MEM_LIMIT_V1); | 130 | std::ifstream memLimitV1(MEM_LIMIT_V1); |
| 116 | std::ifstream memLimitV2(MEM_LIMIT_V2); | 131 | std::ifstream memLimitV2(MEM_LIMIT_V2); |
| @@ -12,6 +12,7 @@ | |||
| 12 | 12 | ||
| 13 | 13 | ||
| 14 | 14 | ||
| 15 | + | ||
| 15 | 16 | ||
| 16 | 17 | ||
| 17 | 18 | ||
| @@ -21,7 +22,6 @@ | |||
| 21 | 22 | ||
| 22 | 23 | ||
| 23 | namespace Adx { | 24 | namespace Adx { |
| 24 | -static const int32_t WAIT_RECORD_FILE_FINISH_TIME = 500; | ||
| 25 | static const std::size_t MAX_IP_LENGTH = 16; | 25 | static const std::size_t MAX_IP_LENGTH = 16; |
| 26 | 26 | ||
| 27 | constexpr char STRING_BIN[] = ".bin"; | 27 | constexpr char STRING_BIN[] = ".bin"; |
| @@ -139,6 +139,11 @@ AdxDumpRecord::AdxDumpRecord() | |||
| 139 | : dumpRecordFlag_(true), | 139 | : dumpRecordFlag_(true), |
| 140 | dumpInitNum_(0) | 140 | dumpInitNum_(0) |
| 141 | { | 141 | { |
| 142 | + int32_t ret = pthread_atfork( | ||
| 143 | + AdxDumpRecord::PrepareFork, AdxDumpRecord::PostForkParent, AdxDumpRecord::PostForkChild); | ||
| 144 | + if (ret != 0) { | ||
| 145 | + IDE_LOGW("call pthread_atfork failed, ret: %d", ret); | ||
| 146 | + } | ||
| 142 | 147 | ||
| 143 | fileNameStatus_.reserve(FILENAME_CHECK_SIZE_MAX); | 148 | fileNameStatus_.reserve(FILENAME_CHECK_SIZE_MAX); |
| 144 | for (size_t idx = 0; idx < FILENAME_CHECK_SIZE_MAX; ++idx) { | 149 | for (size_t idx = 0; idx < FILENAME_CHECK_SIZE_MAX; ++idx) { |
| @@ -186,13 +191,6 @@ bool AdxDumpRecord::CanShutdownServer() const | |||
| 186 | return dumpInitNum_ <= 1; | 191 | return dumpInitNum_ <= 1; |
| 187 | } | 192 | } |
| 188 | 193 | ||
| 189 | -/** | ||
| 190 | - * @brief initialize record file | ||
| 191 | - * @param [in] recordPath : record file Path | ||
| 192 | - * @return | ||
| 193 | - * IDE_DAEMON_ERROR : falied | ||
| 194 | - * IDE_DAEMON_OK : success | ||
| 195 | - */ | ||
| 196 | int32_t AdxDumpRecord::Init(const std::string &hostPid) | 194 | int32_t AdxDumpRecord::Init(const std::string &hostPid) |
| 197 | { | 195 | { |
| 198 | // non-soc case | 196 | // non-soc case |
| @@ -234,13 +232,38 @@ int32_t AdxDumpRecord::Init(const std::string &hostPid) | |||
| 234 | if (!dumpPath_.empty() && dumpPath_.back() != OS_SPLIT_CHAR) { | 232 | if (!dumpPath_.empty() && dumpPath_.back() != OS_SPLIT_CHAR) { |
| 235 | dumpPath_ += OS_SPLIT_CHAR; | 233 | dumpPath_ += OS_SPLIT_CHAR; |
| 236 | } | 234 | } |
| 237 | - hostDumpDataInfoQueue_.Init(); | 235 | + |
| 236 | + std::lock_guard<std::mutex> lock(recordMutex_); | ||
| 237 | + if (hostDumpDataInfoQueue_ == nullptr) { | ||
| 238 | + hostDumpDataInfoQueue_.reset(new(std::nothrow) BoundQueueMemory<HostDumpDataInfo>()); | ||
| 239 | + IDE_CTRL_VALUE_FAILED(hostDumpDataInfoQueue_ != nullptr, return IDE_DAEMON_ERROR, | ||
| 240 | + "Failed to new hostDumpDataInfoQueue"); | ||
| 241 | + } | ||
| 242 | + hostDumpDataInfoQueue_->Init(); | ||
| 238 | IDE_LOGI("record remote dump temp path: %s", dumpPath_.c_str()); | 243 | IDE_LOGI("record remote dump temp path: %s", dumpPath_.c_str()); |
| 239 | - hostDumpDataInfoQueue_.SetPath(dumpPath_); | 244 | + hostDumpDataInfoQueue_->SetPath(dumpPath_); |
| 240 | dumpRecordFlag_ = true; | 245 | dumpRecordFlag_ = true; |
| 241 | return IDE_DAEMON_OK; | 246 | return IDE_DAEMON_OK; |
| 242 | } | 247 | } |
| 243 | 248 | ||
| 249 | +int32_t AdxDumpRecord::StartRecord() | ||
| 250 | +{ | ||
| 251 | + std::lock_guard<std::mutex> lock(recordMutex_); | ||
| 252 | + if (recordThread_.joinable()) { | ||
| 253 | + IDE_LOGI("dump record thread has been started, no need to start again"); | ||
| 254 | + return IDE_DAEMON_OK; | ||
| 255 | + } | ||
| 256 | + dumpRecordFlag_ = true; | ||
| 257 | + try { | ||
| 258 | + recordThread_ = std::thread(&AdxDumpRecord::RecordDumpInfo, this); | ||
| 259 | + } catch (const std::exception &ex) { | ||
| 260 | + dumpRecordFlag_ = false; | ||
| 261 | + IDE_LOGE("Create the dump record thread failed, message: %s", ex.what()); | ||
| 262 | + return IDE_DAEMON_ERROR; | ||
| 263 | + } | ||
| 264 | + return IDE_DAEMON_OK; | ||
| 265 | +} | ||
| 266 | + | ||
| 244 | /** | 267 | /** |
| 245 | * @brief initialize record file | 268 | * @brief initialize record file |
| 246 | * @param [in] recordPath : record file Path | 269 | * @param [in] recordPath : record file Path |
| @@ -250,20 +273,41 @@ int32_t AdxDumpRecord::Init(const std::string &hostPid) | |||
| 250 | */ | 273 | */ |
| 251 | int32_t AdxDumpRecord::UnInit() | 274 | int32_t AdxDumpRecord::UnInit() |
| 252 | { | 275 | { |
| 276 | + std::lock_guard<std::mutex> lock(recordMutex_); | ||
| 253 | IDE_LOGI("start to dump uninit"); | 277 | IDE_LOGI("start to dump uninit"); |
| 254 | dumpRecordFlag_ = false; | 278 | dumpRecordFlag_ = false; |
| 255 | -#if !defined(__IDE_UT) && !defined(__IDE_ST) | 279 | + if (hostDumpDataInfoQueue_ != nullptr) { |
| 256 | - while (!DumpDataQueueIsEmpty()) { | 280 | + hostDumpDataInfoQueue_->Quit(); |
| 257 | - mmSleep(WAIT_RECORD_FILE_FINISH_TIME); | 281 | + } |
| 282 | + if (recordThread_.joinable()) { | ||
| 283 | + recordThread_.join(); | ||
| 258 | } | 284 | } |
| 259 | - | ||
| 260 | - this->hostDumpDataInfoQueue_.Quit(); | ||
| 261 | - | ||
| 262 | - | ||
| 263 | IDE_LOGI("dump uninit success"); | 285 | IDE_LOGI("dump uninit success"); |
| 264 | return IDE_DAEMON_OK; | 286 | return IDE_DAEMON_OK; |
| 265 | } | 287 | } |
| 266 | 288 | ||
| 289 | +void AdxDumpRecord::PrepareFork() | ||
| 290 | +{ | ||
| 291 | + Instance().recordMutex_.lock(); | ||
| 292 | +} | ||
| 293 | + | ||
| 294 | +void AdxDumpRecord::PostForkParent() | ||
| 295 | +{ | ||
| 296 | + Instance().recordMutex_.unlock(); | ||
| 297 | +} | ||
| 298 | + | ||
| 299 | +void AdxDumpRecord::PostForkChild() | ||
| 300 | +{ | ||
| 301 | + auto &instance = Instance(); | ||
| 302 | + instance.dumpRecordFlag_ = false; | ||
| 303 | + if (instance.recordThread_.joinable()) { | ||
| 304 | + instance.recordThread_.detach(); | ||
| 305 | + } | ||
| 306 | + | ||
| 307 | + (void)instance.hostDumpDataInfoQueue_.release(); | ||
| 308 | + instance.recordMutex_.unlock(); | ||
| 309 | +} | ||
| 310 | + | ||
| 267 | /** | 311 | /** |
| 268 | * @brief record dump data to disk | 312 | * @brief record dump data to disk |
| 269 | * @param [in] dumpChunk : dump chunk | 313 | * @param [in] dumpChunk : dump chunk |
| @@ -654,9 +698,9 @@ void AdxDumpRecord::RecordDumpInfo() | |||
| 654 | { | 698 | { |
| 655 | IDE_RUN_LOGI("start dump thread, remote dump record temp path : %s.", dumpPath_.c_str()); | 699 | IDE_RUN_LOGI("start dump thread, remote dump record temp path : %s.", dumpPath_.c_str()); |
| 656 | uint32_t chunkHeaderLen = static_cast<uint32_t>(sizeof(DumpChunk)); | 700 | uint32_t chunkHeaderLen = static_cast<uint32_t>(sizeof(DumpChunk)); |
| 657 | - while (dumpRecordFlag_ || !DumpDataQueueIsEmpty()) { | 701 | + while (hostDumpDataInfoQueue_ != nullptr && (dumpRecordFlag_ || !DumpDataQueueIsEmpty())) { |
| 658 | HostDumpDataInfo data = {nullptr, 0}; | 702 | HostDumpDataInfo data = {nullptr, 0}; |
| 659 | - if (!hostDumpDataInfoQueue_.Pop(data)) { | 703 | + if (!hostDumpDataInfoQueue_->Pop(data)) { |
| 660 | continue; | 704 | continue; |
| 661 | } | 705 | } |
| 662 | 706 | ||
| @@ -674,13 +718,14 @@ void AdxDumpRecord::RecordDumpInfo() | |||
| 674 | "bufLen(%u) exceeds actual data buffer size(%u bytes), fileName: %s", | 718 | "bufLen(%u) exceeds actual data buffer size(%u bytes), fileName: %s", |
| 675 | dumpChunk->bufLen, data.recvLen - chunkHeaderLen, dumpChunk->fileName); | 719 | dumpChunk->bufLen, data.recvLen - chunkHeaderLen, dumpChunk->fileName); |
| 676 | 720 | ||
| 677 | - IDE_LOGI("Queue pop data success! filename: %s, offset: %" PRId64 ", bufLen: %u bytes, isLast: %u, flag: %d.", | 721 | + IDE_LOGI("Queue pop data success! filename: %s, offset: %" PRId64 ", bufLen: %u bytes, " |
| 678 | - dumpChunk->fileName, dumpChunk->offset, dumpChunk->bufLen, dumpChunk->isLastChunk, dumpChunk->flag); | 722 | + "isLast: %u, flag: %d, remaining queue size: %u.", |
| 723 | + dumpChunk->fileName, dumpChunk->offset, dumpChunk->bufLen, dumpChunk->isLastChunk, | ||
| 724 | + dumpChunk->flag, hostDumpDataInfoQueue_->Size()); | ||
| 679 | 725 | ||
 | |||
| 680 | if (FileNameCheck(*dumpChunk)) { | 726 | if (FileNameCheck(*dumpChunk)) { |
| 681 | IDE_CTRL_VALUE_FAILED_NODO(StatsDataParsing(*dumpChunk), continue, | 727 | IDE_CTRL_VALUE_FAILED_NODO(StatsDataParsing(*dumpChunk), continue, |
| 682 | "Failed to parse dump data with file name %s", dumpChunk->fileName); | 728 | "Failed to parse dump data with file name %s", dumpChunk->fileName); |
| 683 | - IDE_LOGD("New popped data process success"); | ||
| 684 | continue; | 729 | continue; |
| 685 | } | 730 | } |
| 686 | 731 | ||
| @@ -701,7 +746,6 @@ void AdxDumpRecord::RecordDumpInfo() | |||
| 701 | } else if (!RecordDumpDataToDisk(*dumpChunk)) { | 746 | } else if (!RecordDumpDataToDisk(*dumpChunk)) { |
| 702 | IDE_LOGE("failed to record dump data to disk."); | 747 | IDE_LOGE("failed to record dump data to disk."); |
| 703 | } | 748 | } |
| 704 | - IDE_LOGD("new popped data process success"); | ||
| 705 | } | 749 | } |
| 706 | IDE_LOGI("exit record file thread"); | 750 | IDE_LOGI("exit record file thread"); |
| 707 | } | 751 | } |
| @@ -715,15 +759,22 @@ void AdxDumpRecord::RecordDumpInfo() | |||
| 715 | */ | 759 | */ |
| 716 | bool AdxDumpRecord::RecordDumpDataToQueue(HostDumpDataInfo &info) | 760 | bool AdxDumpRecord::RecordDumpDataToQueue(HostDumpDataInfo &info) |
| 717 | { | 761 | { |
| 718 | - if (hostDumpDataInfoQueue_.IsFull()) { | 762 | + if (hostDumpDataInfoQueue_ == nullptr) { |
| 763 | + IDE_LOGW("dump data queue is not initialized, drop this data packet."); | ||
| 764 | + return false; | ||
| 765 | + } | ||
| 766 | + if (hostDumpDataInfoQueue_->IsFull()) { | ||
| 719 | const std::string tipFull = "Memory usage exceeds 85%, the dump data queue is full"; | 767 | const std::string tipFull = "Memory usage exceeds 85%, the dump data queue is full"; |
| 720 | const std::string tipReduce = "Please reduce model batches, images or dump layers"; | 768 | const std::string tipReduce = "Please reduce model batches, images or dump layers"; |
| 721 | const std::string tipMemory = "Or clear the used memory or increase the maximum memory"; | 769 | const std::string tipMemory = "Or clear the used memory or increase the maximum memory"; |
| 722 | IDE_LOGW("%s. %s. %s.", tipFull.c_str(), tipReduce.c_str(), tipMemory.c_str()); | 770 | IDE_LOGW("%s. %s. %s.", tipFull.c_str(), tipReduce.c_str(), tipMemory.c_str()); |
| 723 | return false; | 771 | return false; |
| 724 | } else { | 772 | } else { |
| 725 | - hostDumpDataInfoQueue_.Push(info); | 773 | + if (!hostDumpDataInfoQueue_->Push(info)) { |
| 726 | - IDE_LOGD("Insert dump data to queue success."); | 774 | + IDE_LOGW("dump data queue has quit, drop this data packet."); |
| 775 | + return false; | ||
| 776 | + } | ||
| 777 | + IDE_LOGI("Insert dump data to queue success, queue size: %u.", hostDumpDataInfoQueue_->Size()); | ||
| 727 | } | 778 | } |
| 728 | 779 | ||
| 729 | return true; | 780 | return true; |
| @@ -738,7 +789,7 @@ bool AdxDumpRecord::RecordDumpDataToQueue(HostDumpDataInfo &info) | |||
| 738 | */ | 789 | */ |
| 739 | bool AdxDumpRecord::DumpDataQueueIsEmpty() const | 790 | bool AdxDumpRecord::DumpDataQueueIsEmpty() const |
| 740 | { | 791 | { |
| 741 | - return hostDumpDataInfoQueue_.IsEmpty(); | 792 | + return hostDumpDataInfoQueue_ == nullptr || hostDumpDataInfoQueue_->IsEmpty(); |
| 742 | } | 793 | } |
| 743 | 794 | ||
| 744 | void AdxDumpRecord::SetDumpPath(const std::string &dumpPath) | 795 | void AdxDumpRecord::SetDumpPath(const std::string &dumpPath) |
| @@ -9,10 +9,13 @@ | |||
| 9 | */ | 9 | */ |
| 10 | 10 | ||
| 11 | 11 | ||
| 12 | + | ||
| 13 | + | ||
| 12 | 14 | ||
| 13 | 15 | ||
| 14 | 16 | ||
| 15 | 17 | ||
| 18 | + | ||
| 16 | 19 | ||
| 17 | 20 | ||
| 18 | 21 | ||
| @@ -84,6 +87,7 @@ public: | |||
| 84 | int32_t Init(const std::string &hostPid); | 87 | int32_t Init(const std::string &hostPid); |
| 85 | int32_t UnInit(); | 88 | int32_t UnInit(); |
| 86 | void SetWorkPath(const std::string &path); | 89 | void SetWorkPath(const std::string &path); |
| 90 | + int32_t StartRecord(); | ||
| 87 | void RecordDumpInfo(); | 91 | void RecordDumpInfo(); |
| 88 | bool RecordDumpDataToQueue(HostDumpDataInfo &info); | 92 | bool RecordDumpDataToQueue(HostDumpDataInfo &info); |
| 89 | bool DumpDataQueueIsEmpty() const; | 93 | bool DumpDataQueueIsEmpty() const; |
| @@ -97,6 +101,9 @@ public: | |||
| 97 | 101 | ||
| 98 | private: | 102 | private: |
| 99 | bool JudgeRemoteFalg(const std::string &msg) const; | 103 | bool JudgeRemoteFalg(const std::string &msg) const; |
| 104 | + static void PrepareFork(); | ||
| 105 | + static void PostForkParent(); | ||
| 106 | + static void PostForkChild(); | ||
| 100 | 107 | ||
| 101 | bool DumpDataToCallback(const std::string &filename, const std::string &dumpData, int64_t offSet, int32_t flag); | 108 | bool DumpDataToCallback(const std::string &filename, const std::string &dumpData, int64_t offSet, int32_t flag); |
| 102 | bool StatsDataParsing(const DumpChunk &dumpChunk); | 109 | bool StatsDataParsing(const DumpChunk &dumpChunk); |
| @@ -117,11 +124,14 @@ private: | |||
| 117 | 124 | ||
| 118 | 125 | ||
| 119 | private: | 126 | private: |
| 120 | - bool dumpRecordFlag_; | 127 | + std::atomic<bool> dumpRecordFlag_; |
| 121 | std::string dumpPath_; | 128 | std::string dumpPath_; |
| 122 | std::string workPath_; | 129 | std::string workPath_; |
| 123 | - BoundQueueMemory<HostDumpDataInfo> hostDumpDataInfoQueue_; | 130 | + std::unique_ptr<BoundQueueMemory<HostDumpDataInfo>> hostDumpDataInfoQueue_; |
| 124 | int32_t dumpInitNum_; | 131 | int32_t dumpInitNum_; |
| 132 | + std::thread recordThread_; | ||
| 133 | + std::mutex recordMutex_; | ||
| 134 | + | ||
| 125 | 135 | ||
| 126 | uint64_t dumpStatsItem_{0}; | 136 | uint64_t dumpStatsItem_{0}; |
| 127 | uint32_t filenameIndex_{0}; | 137 | uint32_t filenameIndex_{0}; |
| @@ -22,13 +22,29 @@ constexpr int64_t MAX_BUFFER_LENGTH = 512; | |||
| 22 | static const std::string MAPPING_FILE_NAME = "mapping.csv"; | 22 | static const std::string MAPPING_FILE_NAME = "mapping.csv"; |
| 23 | } // namespace | 23 | } // namespace |
| 24 | 24 | ||
| 25 | -File::File(const std::string &path, int32_t flag, mmMode_t mode) : filePath_(path), fd_(INVALID_FILE_FD) | 25 | +File::File(const std::string &path, int32_t flag, mmMode_t mode, bool lazyOpen) |
| 26 | + : filePath_(path), fd_(INVALID_FILE_FD), flag_(flag), mode_(mode) | ||
| 26 | { | 27 | { |
| 28 | + if (lazyOpen) { | ||
| 29 | + return; | ||
| 30 | + } | ||
| 27 | if (Open(flag, mode) != ADUMP_SUCCESS) { | 31 | if (Open(flag, mode) != ADUMP_SUCCESS) { |
| 28 | fd_ = INVALID_FILE_FD; | 32 | fd_ = INVALID_FILE_FD; |
| 29 | } | 33 | } |
| 30 | } | 34 | } |
| 31 | 35 | ||
| 36 | +int32_t File::EnsureOpen() | ||
| 37 | +{ | ||
| 38 | + if (fd_ != INVALID_FILE_FD) { | ||
| 39 | + return ADUMP_SUCCESS; | ||
| 40 | + } | ||
| 41 | + if (Open(flag_, mode_) != ADUMP_SUCCESS) { | ||
| 42 | + fd_ = INVALID_FILE_FD; | ||
| 43 | + return ADUMP_FAILED; | ||
| 44 | + } | ||
| 45 | + return ADUMP_SUCCESS; | ||
| 46 | +} | ||
| 47 | + | ||
| 32 | File::~File() | 48 | File::~File() |
| 33 | { | 49 | { |
| 34 | (void)Close(); | 50 | (void)Close(); |
| @@ -16,9 +16,10 @@ | |||
| 16 | namespace Adx { | 16 | namespace Adx { |
| 17 | class File { | 17 | class File { |
| 18 | public: | 18 | public: |
| 19 | - explicit File(const std::string &path, int32_t flag, mmMode_t mode = M_IRUSR | M_IWUSR); | 19 | + explicit File(const std::string &path, int32_t flag, mmMode_t mode = M_IRUSR | M_IWUSR, bool lazyOpen = false); |
| 20 | ~File(); | 20 | ~File(); |
| 21 | int32_t IsFileOpen() const; | 21 | int32_t IsFileOpen() const; |
| 22 | + int32_t EnsureOpen(); | ||
| 22 | int64_t Write(const char * const buffer, int64_t length) const; | 23 | int64_t Write(const char * const buffer, int64_t length) const; |
| 23 | int64_t Read(char *buffer, int64_t length) const; | 24 | int64_t Read(char *buffer, int64_t length) const; |
| 24 | static int32_t Copy(const std::string &srcPath, const std::string &dstPath); | 25 | static int32_t Copy(const std::string &srcPath, const std::string &dstPath); |
| @@ -29,6 +30,8 @@ private: | |||
| 29 | int32_t AddMapping(const std::string &filePath, const std::string &fileName, const std::string &hashName); | 30 | int32_t AddMapping(const std::string &filePath, const std::string &fileName, const std::string &hashName); |
| 30 | std::string filePath_; | 31 | std::string filePath_; |
| 31 | int32_t fd_; | 32 | int32_t fd_; |
| 33 | + int32_t flag_; | ||
| 34 | + mmMode_t mode_; | ||
| 32 | }; | 35 | }; |
| 33 | } // namespace Adx | 36 | } // namespace Adx |
| 34 | 37 | ||
| @@ -15,43 +15,17 @@ | |||
| 15 | 15 | ||
| 16 | 16 | ||
| 17 | namespace Adx { | 17 | namespace Adx { |
| 18 | -static IdeThreadArg AdxDumpRecordSocProcess(const IdeThreadArg arg) | ||
| 19 | -{ | ||
| 20 | - UNUSED(arg); | ||
| 21 | - AdxDumpRecord::Instance().RecordDumpInfo(); | ||
| 22 | - return nullptr; | ||
| 23 | -} | ||
| 24 | - | ||
| 25 | -static IdeThreadArg AdxDataDumpServerSocProcess(const IdeThreadArg arg) | ||
| 26 | -{ | ||
| 27 | - UNUSED(arg); | ||
| 28 | - mmUserBlock_t funcBlock; | ||
| 29 | - funcBlock.pulArg = nullptr; | ||
| 30 | - mmThread tid = 0; | ||
| 31 | - funcBlock.procFunc = AdxDumpRecordSocProcess; | ||
| 32 | - int ret = Thread::CreateTaskWithDefaultAttr(tid, funcBlock); | ||
| 33 | - if (ret != EN_OK) { | ||
| 34 | - return nullptr; | ||
| 35 | - } | ||
| 36 | - | ||
| 37 | - (void)mmJoinTask(&tid); | ||
| 38 | - return nullptr; | ||
| 39 | -} | ||
| 40 | - | ||
| 41 | int32_t AdxSocDataDumpInit(const std::string &hostPid) | 18 | int32_t AdxSocDataDumpInit(const std::string &hostPid) |
| 42 | { | 19 | { |
| 43 | - mmUserBlock_t funcBlock; | ||
| 44 | - funcBlock.procFunc = AdxDataDumpServerSocProcess; | ||
| 45 | - funcBlock.pulArg = nullptr; | ||
| 46 | - mmThread tid = 0; | ||
| 47 | // soc case, pass host pid to record instance | 20 | // soc case, pass host pid to record instance |
| 48 | int ret = AdxDumpRecord::Instance().Init(hostPid); | 21 | int ret = AdxDumpRecord::Instance().Init(hostPid); |
| 49 | if (ret != IDE_DAEMON_OK) { | 22 | if (ret != IDE_DAEMON_OK) { |
| 50 | IDE_LOGE("AdxDumpRecord init failed."); | 23 | IDE_LOGE("AdxDumpRecord init failed."); |
| 51 | return IDE_DAEMON_ERROR; | 24 | return IDE_DAEMON_ERROR; |
| 52 | } | 25 | } |
| 53 | - ret = Thread::CreateDetachTaskWithDefaultAttr(tid, funcBlock); | 26 | + ret = AdxDumpRecord::Instance().StartRecord(); |
| 54 | - if (ret != EN_OK) { | 27 | + if (ret != IDE_DAEMON_OK) { |
| 28 | + IDE_LOGE("start dump record thread failed."); | ||
| 55 | return IDE_DAEMON_ERROR; | 29 | return IDE_DAEMON_ERROR; |
| 56 | } | 30 | } |
| 57 | IDE_LOGI("Adx soc dump thread has been started."); | 31 | IDE_LOGI("Adx soc dump thread has been started."); |
| @@ -218,9 +218,27 @@ void DumpFile::SetTensorBuffer(const std::vector<TensorBuffer> &tensorBuffer) | |||
| 218 | } | 218 | } |
| 219 | } | 219 | } |
| 220 | 220 | ||
| 221 | +bool DumpFile::HasDumpData() const | ||
| 222 | +{ | ||
| 223 | + if (!inputs_.empty() || !outputs_.empty() || !workspaces_.empty() || !inputBuffer_.empty()) { | ||
| 224 | + return true; | ||
| 225 | + } | ||
| 226 | + | ||
| 227 | + if (!mc2Spaces_.empty()) { | ||
| 228 | + return true; | ||
| 229 | + } | ||
| 230 | + | ||
| 231 | + return false; | ||
| 232 | +} | ||
| 233 | + | ||
| 221 | int32_t DumpFile::Dump(std::vector<std::string> &record) | 234 | int32_t DumpFile::Dump(std::vector<std::string> &record) |
| 222 | { | 235 | { |
| 223 | - int32_t ret = file_.IsFileOpen(); | 236 | + if (!HasDumpData()) { |
| 237 | + IDE_LOGI("No dump data, skip dump file."); | ||
| 238 | + return ADUMP_SUCCESS; | ||
| 239 | + } | ||
| 240 | + | ||
| 241 | + int32_t ret = file_.EnsureOpen(); | ||
| 224 | if (ret != ADUMP_SUCCESS) { | 242 | if (ret != ADUMP_SUCCESS) { |
| 225 | return ret; | 243 | return ret; |
| 226 | } | 244 | } |
| @@ -24,7 +24,7 @@ namespace Adx { | |||
| 24 | class DumpFile { | 24 | class DumpFile { |
| 25 | public: | 25 | public: |
| 26 | DumpFile(const uint32_t deviceId, const std::string &filePath) : deviceId_(deviceId), | 26 | DumpFile(const uint32_t deviceId, const std::string &filePath) : deviceId_(deviceId), |
| 27 | - file_(filePath, M_RDWR | M_CREAT | M_APPEND, M_IRUSR | M_IWUSR) {} | 27 | + file_(filePath, M_RDWR | M_CREAT | M_APPEND, M_IRUSR | M_IWUSR, true) {} |
| 28 | void SetHeader(const std::string &opName); | 28 | void SetHeader(const std::string &opName); |
| 29 | void SetInputTensors(const std::vector<DumpTensor> &inputTensors); | 29 | void SetInputTensors(const std::vector<DumpTensor> &inputTensors); |
| 30 | void SetOutputTensors(const std::vector<DumpTensor> &outputTensors); | 30 | void SetOutputTensors(const std::vector<DumpTensor> &outputTensors); |
| @@ -37,6 +37,7 @@ public: | |||
| 37 | 37 | ||
| 38 | 38 | ||
| 39 | private: | 39 | private: |
| 40 | + bool HasDumpData() const; | ||
| 40 | void SetAicInfo(std::vector<std::string> &record); | 41 | void SetAicInfo(std::vector<std::string> &record); |
| 41 | int32_t WriteHeader() const; | 42 | int32_t WriteHeader() const; |
| 42 | int32_t WriteInputTensors(); | 43 | int32_t WriteInputTensors(); |
| @@ -43,13 +43,6 @@ IdeThreadArg AdxDataDumpServerProcess(const IdeThreadArg arg) | |||
| 43 | return nullptr; | 43 | return nullptr; |
| 44 | } | 44 | } |
| 45 | 45 | ||
| 46 | -IdeThreadArg AdxDumpRecordProcess(const IdeThreadArg arg) | ||
| 47 | -{ | ||
| 48 | - UNUSED(arg); | ||
| 49 | - AdxDumpRecord::Instance().RecordDumpInfo(); | ||
| 50 | - return nullptr; | ||
| 51 | -} | ||
| 52 | - | ||
| 53 | static bool IsOnDeviceSide() | 46 | static bool IsOnDeviceSide() |
| 54 | { | 47 | { |
| 55 | uint32_t platformInfo = static_cast<uint32_t>(SysPlatformType::INVALID); | 48 | uint32_t platformInfo = static_cast<uint32_t>(SysPlatformType::INVALID); |
| @@ -77,10 +70,9 @@ int32_t AdxDataDumpServerInit() | |||
| 77 | IDE_LOGE("AdxDumpRecord init failed."); | 70 | IDE_LOGE("AdxDumpRecord init failed."); |
| 78 | return IDE_DAEMON_ERROR; | 71 | return IDE_DAEMON_ERROR; |
| 79 | } | 72 | } |
| 80 | - funcBlock.procFunc = AdxDumpRecordProcess; | 73 | + ret = AdxDumpRecord::Instance().StartRecord(); |
| 81 | - funcBlock.pulArg = nullptr; | 74 | + if (ret != IDE_DAEMON_OK) { |
| 82 | - ret = Thread::CreateDetachTaskWithDefaultAttr(tid, funcBlock); | 75 | + IDE_LOGE("start dump record thread failed."); |
| 83 | - if (ret != EN_OK) { | ||
| 84 | return IDE_DAEMON_ERROR; | 76 | return IDE_DAEMON_ERROR; |
| 85 | } | 77 | } |
| 86 | 78 | ||
| @@ -110,23 +102,25 @@ int32_t AdxDataDumpServerUnInit() | |||
| 110 | } | 102 | } |
| 111 | 103 | ||
| 112 | IDE_LOGI("start to do dump uninit"); | 104 | IDE_LOGI("start to do dump uninit"); |
| 113 | - if (AdxDumpRecord::Instance().UnInit() != IDE_DAEMON_OK) { | ||
| 114 | - IDE_LOGE("dump record uninit failed"); | ||
| 115 | - return IDE_DAEMON_ERROR; | ||
| 116 | - } | ||
| 117 | - | ||
| 118 | std::string hostPid; | 105 | std::string hostPid; |
| 119 | ADX_GET_ENV(MM_ENV_ASCEND_HOSTPID, hostPid); | 106 | ADX_GET_ENV(MM_ENV_ASCEND_HOSTPID, hostPid); |
| 120 | - if (IsOnDeviceSide() && !hostPid.empty()) { | 107 | + int32_t ret = IDE_DAEMON_OK; |
| 121 | - AdxDumpRecord::Instance().UpdateDumpInitNum(false); | 108 | + bool isHelper = IsOnDeviceSide() && !hostPid.empty(); |
| 122 | - IDE_LOGI("dump server not start on helper device"); | 109 | + if (!isHelper && g_manager.Exit() != IDE_DAEMON_OK) { |
| 123 | - return IDE_DAEMON_OK; | ||
| 124 | - } | ||
| 125 | - if (g_manager.Exit() != IDE_DAEMON_OK) { | ||
| 126 | IDE_LOGE("AdxServerManager Exit failed"); | 110 | IDE_LOGE("AdxServerManager Exit failed"); |
| 127 | - return IDE_DAEMON_ERROR; | 111 | + ret = IDE_DAEMON_ERROR; |
| 112 | + } | ||
| 113 | + if (isHelper) { | ||
| 114 | + IDE_LOGI("dump server not start on helper device"); | ||
| 128 | } | 115 | } |
| 129 | 116 | ||
| 130 | - AdxDumpRecord::Instance().UpdateDumpInitNum(false); | 117 | + if (AdxDumpRecord::Instance().UnInit() != IDE_DAEMON_OK) { |
| 131 | - return IDE_DAEMON_OK; | 118 | + IDE_LOGE("dump record uninit failed"); |
| 119 | + ret = IDE_DAEMON_ERROR; | ||
| 120 | + } | ||
| 121 | + | ||
| 122 | + if (ret == IDE_DAEMON_OK) { | ||
| 123 | + AdxDumpRecord::Instance().UpdateDumpInitNum(false); | ||
| 124 | + } | ||
| 125 | + return ret; | ||
| 132 | } | 126 | } |
| @@ -172,6 +172,7 @@ int32_t KernelDfxDumper::UnInitTask() | |||
| 172 | void KernelDfxDumper::UnInit() | 172 | void KernelDfxDumper::UnInit() |
| 173 | { | 173 | { |
| 174 | std::lock_guard<std::mutex> lock(mutex_); | 174 | std::lock_guard<std::mutex> lock(mutex_); |
| 175 | + destructed_ = true; | ||
| 175 | UnInitTask(); | 176 | UnInitTask(); |
| 176 | dumpPath_.clear(); | 177 | dumpPath_.clear(); |
| 177 | enabledDfxTypes_.clear(); | 178 | enabledDfxTypes_.clear(); |
| @@ -207,7 +208,6 @@ void KernelDfxDumper::PostForkChild() | |||
| 207 | instance.dumpDfxInfoQueue_ = nullptr; | 208 | instance.dumpDfxInfoQueue_ = nullptr; |
| 208 | } | 209 | } |
| 209 | instance.mutex_.unlock(); | 210 | instance.mutex_.unlock(); |
| 210 | - IDE_LOGI("KernelDfxDumper has been reset in the child process."); | ||
| 211 | } | 211 | } |
| 212 | 212 | ||
| 213 | KernelDfxDumper::KernelDfxDumper() | 213 | KernelDfxDumper::KernelDfxDumper() |
| @@ -263,6 +263,7 @@ int32_t KernelDfxDumper::EnableDfxDumper(const DumpDfxConfig config) | |||
| 263 | return ADUMP_SUCCESS; | 263 | return ADUMP_SUCCESS; |
| 264 | } | 264 | } |
| 265 | std::lock_guard<std::mutex> lock(mutex_); | 265 | std::lock_guard<std::mutex> lock(mutex_); |
| 266 | + destructed_ = false; | ||
| 266 | std::set<rtKernelDfxInfoType> rtDfxTypes; | 267 | std::set<rtKernelDfxInfoType> rtDfxTypes; |
| 267 | GetRegisterDfxTypes(config.dfxTypes, rtDfxTypes); | 268 | GetRegisterDfxTypes(config.dfxTypes, rtDfxTypes); |
| 268 | for (auto& rtDfxType : rtDfxTypes) { | 269 | for (auto& rtDfxType : rtDfxTypes) { |
| @@ -326,6 +327,9 @@ std::string KernelDfxDumper::GetDfxInfoFilePath(uint32_t coreId, std::string &co | |||
| 326 | int32_t KernelDfxDumper::DumpKernelDfxInfo(rtKernelDfxInfoType dfxType, uint32_t coreType, uint32_t coreId, | 327 | int32_t KernelDfxDumper::DumpKernelDfxInfo(rtKernelDfxInfoType dfxType, uint32_t coreType, uint32_t coreId, |
| 327 | const uint8_t *buffer, size_t length) | 328 | const uint8_t *buffer, size_t length) |
| 328 | { | 329 | { |
| 330 | + std::lock_guard<std::mutex> lock(mutex_); | ||
| 331 | + IDE_CTRL_VALUE_WARN(!destructed_, return ADUMP_FAILED, "KernelDfxDumper has been destructed."); | ||
| 332 | + | ||
| 329 | std::string dfxTypeStr = GetDfxTypeStr(dfxType); | 333 | std::string dfxTypeStr = GetDfxTypeStr(dfxType); |
| 330 | std::string coreTypeStr = GetCoreTypeStr(coreType); | 334 | std::string coreTypeStr = GetCoreTypeStr(coreType); |
| 331 | IDE_CTRL_VALUE_WARN(IsEnabled(dfxType), return ADUMP_FAILED, | 335 | IDE_CTRL_VALUE_WARN(IsEnabled(dfxType), return ADUMP_FAILED, |
| @@ -59,6 +59,7 @@ private: | |||
| 59 | std::string dumpPath_; | 59 | std::string dumpPath_; |
| 60 | std::atomic<bool> taskInit_ = false; | 60 | std::atomic<bool> taskInit_ = false; |
| 61 | std::atomic<bool> taskRunning_ = false; | 61 | std::atomic<bool> taskRunning_ = false; |
| 62 | + bool destructed_ = false; | ||
| 62 | std::thread taskThread_; | 63 | std::thread taskThread_; |
| 63 | std::mutex mutex_; | 64 | std::mutex mutex_; |
| 64 | std::unique_ptr<BoundQueueMemory<DumpDfxInfo>> dumpDfxInfoQueue_; | 65 | std::unique_ptr<BoundQueueMemory<DumpDfxInfo>> dumpDfxInfoQueue_; |
| @@ -492,15 +492,16 @@ TEST_F(DumpArgsCallbackUtest, Test_Constructor_WithDisplayName) | |||
| 492 | Tools::CaseWorkspace ws("Test_Constructor_WithDisplayName"); | 492 | Tools::CaseWorkspace ws("Test_Constructor_WithDisplayName"); |
| 493 | rtExceptionInfo exception = {0}; | 493 | rtExceptionInfo exception = {0}; |
| 494 | InitExceptionInfo(exception); | 494 | InitExceptionInfo(exception); |
| 495 | - | 495 | + |
| 496 | ExceptionDumpInfo info = {0}; | 496 | ExceptionDumpInfo info = {0}; |
| 497 | info.coreId = 0; | 497 | info.coreId = 0; |
| 498 | info.coreType = 1; | 498 | info.coreType = 1; |
| 499 | SetKernelName(info, "test"); | 499 | SetKernelName(info, "test"); |
| 500 | SetDisplayName(info, "display"); | 500 | SetDisplayName(info, "display"); |
| 501 | 501 | ||
| 502 | + // 无 dfx args / extra tensor / 日志时，Dump 应跳过并返回成功，且不生成空文件 | ||
| 502 | DumpArgsCallback callback(exception, info, ws.Root()); | 503 | DumpArgsCallback callback(exception, info, ws.Root()); |
| 503 | - EXPECT_EQ(callback.Dump(), ADUMP_FAILED); | 504 | + EXPECT_EQ(callback.Dump(), ADUMP_SUCCESS); |
| 504 | } | 505 | } |
| 505 | 506 | ||
| 506 | TEST_F(DumpArgsCallbackUtest, Test_Constructor_EmptyDisplayName) | 507 | TEST_F(DumpArgsCallbackUtest, Test_Constructor_EmptyDisplayName) |
| @@ -508,13 +509,14 @@ TEST_F(DumpArgsCallbackUtest, Test_Constructor_EmptyDisplayName) | |||
| 508 | Tools::CaseWorkspace ws("Test_Constructor_EmptyDisplayName"); | 509 | Tools::CaseWorkspace ws("Test_Constructor_EmptyDisplayName"); |
| 509 | rtExceptionInfo exception = {0}; | 510 | rtExceptionInfo exception = {0}; |
| 510 | InitExceptionInfo(exception); | 511 | InitExceptionInfo(exception); |
| 511 | - | 512 | + |
| 512 | ExceptionDumpInfo info = {0}; | 513 | ExceptionDumpInfo info = {0}; |
| 513 | SetKernelName(info, "test"); | 514 | SetKernelName(info, "test"); |
| 514 | info.kernelDisplayName[0] = '\0'; | 515 | info.kernelDisplayName[0] = '\0'; |
| 515 | - | 516 | + |
| 517 | + // 无数据无日志的空 dump 跳过，返回成功且不落盘 | ||
| 516 | DumpArgsCallback callback(exception, info, ws.Root()); | 518 | DumpArgsCallback callback(exception, info, ws.Root()); |
| 517 | - EXPECT_EQ(callback.Dump(), ADUMP_FAILED); | 519 | + EXPECT_EQ(callback.Dump(), ADUMP_SUCCESS); |
| 518 | } | 520 | } |
| 519 | 521 | ||
| 520 | TEST_F(DumpArgsCallbackUtest, Test_DumpExtraTensors_SkipNull) | 522 | TEST_F(DumpArgsCallbackUtest, Test_DumpExtraTensors_SkipNull) |
| @@ -92,14 +92,10 @@ TEST_F(DumpFileUtest, Test_DumpData) | |||
| 92 | int32_t ret = dumpFile.Dump(logRecord); | 92 | int32_t ret = dumpFile.Dump(logRecord); |
| 93 | EXPECT_EQ(ret, ADUMP_SUCCESS); | 93 | EXPECT_EQ(ret, ADUMP_SUCCESS); |
| 94 | 94 | ||
| 95 | + // 所有 tensorAddr 均为 nullptr，无任何真实 tensor/workspace/buffer 数据进入落盘容器， | ||
| 96 | + // 日志仅为附加信息；新口径下无真实数据不落盘，故文件不应生成 | ||
| 95 | DumpFileChecker checker; | 97 | DumpFileChecker checker; |
| 96 | - EXPECT_EQ(checker.Load(dumpFilePath), true); | 98 | + EXPECT_EQ(checker.Load(dumpFilePath), false); |
| 97 | - EXPECT_EQ(checker.CheckHead("test_op"), true); | ||
| 98 | - // tensorAddr is nullptr, so SetInputTensors/SetOutputTensors AND SetWorkspaces skip it; | ||
| 99 | - // only the header is written | ||
| 100 | - EXPECT_EQ(checker.CheckInputTensorNum(0), true); | ||
| 101 | - EXPECT_EQ(checker.CheckOutputTensorNum(0), true); | ||
| 102 | - EXPECT_EQ(checker.CheckWorkspaceNum(0), true); | ||
| 103 | } | 99 | } |
| 104 | 100 | ||
| 105 | // TEST_F(DumpFileUtest, Test_Dump_With_CopyDeviceData_Fail) | 101 | // TEST_F(DumpFileUtest, Test_Dump_With_CopyDeviceData_Fail) |
| @@ -9,6 +9,8 @@ | |||
| 9 | */ | 9 | */ |
| 10 | 10 | ||
| 11 | 11 | ||
| 12 | + | ||
| 13 | + | ||
| 12 | 14 | ||
| 13 | 15 | ||
| 14 | 16 | ||
| @@ -85,6 +87,109 @@ TEST_F(ADX_DUMP_RECORD_TEST, UnInit) | |||
| 85 | EXPECT_EQ(IDE_DAEMON_OK, ret); | 87 | EXPECT_EQ(IDE_DAEMON_OK, ret); |
| 86 | } | 88 | } |
| 87 | 89 | ||
| 90 | +TEST_F(ADX_DUMP_RECORD_TEST, StartRecordAndUnInit) | ||
| 91 | +{ | ||
| 92 | + // start the consumer thread, then UnInit drains the queue and joins it | ||
| 93 | + Adx::AdxDumpRecord::Instance().Init(""); | ||
| 94 | + EXPECT_EQ(IDE_DAEMON_OK, Adx::AdxDumpRecord::Instance().StartRecord()); | ||
| 95 | + // starting again while running is a no-op success | ||
| 96 | + EXPECT_EQ(IDE_DAEMON_OK, Adx::AdxDumpRecord::Instance().StartRecord()); | ||
| 97 | + | ||
| 98 | + const char *srcFile = "adx_data_dump_server_manager"; | ||
| 99 | + uint32_t dataLen = strlen(srcFile) + 1 + sizeof(Adx::DumpChunk); | ||
| 100 | + MsgProto *msg = Adx::AdxMsgProto::CreateMsgPacket(IDE_DUMP_REQ, 0, nullptr, dataLen); | ||
| 101 | + Adx::SharedPtr<MsgProto> msgPtr(msg, IdeXfree); | ||
| 102 | + Adx::HostDumpDataInfo info = {msgPtr, dataLen}; | ||
| 103 | + EXPECT_EQ(true, Adx::AdxDumpRecord::Instance().RecordDumpDataToQueue(info)); | ||
| 104 | + | ||
| 105 | + EXPECT_EQ(IDE_DAEMON_OK, Adx::AdxDumpRecord::Instance().UnInit()); | ||
| 106 | + // after join, the queue must be fully drained | ||
| 107 | + EXPECT_EQ(true, Adx::AdxDumpRecord::Instance().DumpDataQueueIsEmpty()); | ||
| 108 | +} | ||
| 109 | + | ||
| 110 | +// queue is released after fork in child / before Init; access must be null-safe (no crash) | ||
| 111 | +TEST_F(ADX_DUMP_RECORD_TEST, QueueNullSafeAfterRelease) | ||
| 112 | +{ | ||
| 113 | + Adx::AdxDumpRecord::Instance().Init(""); | ||
| 114 | + EXPECT_NE(nullptr, Adx::AdxDumpRecord::Instance().hostDumpDataInfoQueue_.get()); | ||
| 115 | + // simulate the post-fork-child release: drop ownership of the inherited queue | ||
| 116 | + (void)Adx::AdxDumpRecord::Instance().hostDumpDataInfoQueue_.release(); | ||
| 117 | + EXPECT_EQ(nullptr, Adx::AdxDumpRecord::Instance().hostDumpDataInfoQueue_.get()); | ||
| 118 | + | ||
| 119 | + // null queue must be treated as empty so the consumer loop can exit | ||
| 120 | + EXPECT_EQ(true, Adx::AdxDumpRecord::Instance().DumpDataQueueIsEmpty()); | ||
| 121 | + | ||
| 122 | + // enqueue on a null queue must fail gracefully instead of crashing | ||
| 123 | + const char *srcFile = "adx_data_dump_server_manager"; | ||
| 124 | + uint32_t dataLen = strlen(srcFile) + 1 + sizeof(Adx::DumpChunk); | ||
| 125 | + MsgProto *msg = Adx::AdxMsgProto::CreateMsgPacket(IDE_DUMP_REQ, 0, nullptr, dataLen); | ||
| 126 | + Adx::SharedPtr<MsgProto> msgPtr(msg, IdeXfree); | ||
| 127 | + Adx::HostDumpDataInfo info = {msgPtr, dataLen}; | ||
| 128 | + EXPECT_EQ(false, Adx::AdxDumpRecord::Instance().RecordDumpDataToQueue(info)); | ||
| 129 | + | ||
| 130 | + // RecordDumpInfo on a null queue must return immediately, not spin | ||
| 131 | + Adx::AdxDumpRecord::Instance().dumpRecordFlag_ = true; | ||
| 132 | + Adx::AdxDumpRecord::Instance().RecordDumpInfo(); | ||
| 133 | + | ||
| 134 | + // a fresh Init rebuilds the queue so dump can work again | ||
| 135 | + EXPECT_EQ(IDE_DAEMON_OK, Adx::AdxDumpRecord::Instance().Init("")); | ||
| 136 | + EXPECT_NE(nullptr, Adx::AdxDumpRecord::Instance().hostDumpDataInfoQueue_.get()); | ||
| 137 | +} | ||
| 138 | + | ||
| 139 | +// directly drive the pthread_atfork callbacks: child must reset to a clean, restartable state | ||
| 140 | +TEST_F(ADX_DUMP_RECORD_TEST, ForkCallbacksResetChildState) | ||
| 141 | +{ | ||
| 142 | + Adx::AdxDumpRecord::Instance().Init(""); | ||
| 143 | + EXPECT_EQ(IDE_DAEMON_OK, Adx::AdxDumpRecord::Instance().StartRecord()); | ||
| 144 | + | ||
| 145 | + // PrepareFork locks recordMutex_; PostForkParent unlocks it (parent path keeps queue & thread) | ||
| 146 | + Adx::AdxDumpRecord::PrepareFork(); | ||
| 147 | + Adx::AdxDumpRecord::PostForkParent(); | ||
| 148 | + EXPECT_NE(nullptr, Adx::AdxDumpRecord::Instance().hostDumpDataInfoQueue_.get()); | ||
| 149 | + | ||
| 150 | + // parent still owns a joinable thread; reclaim it cleanly | ||
| 151 | + EXPECT_EQ(IDE_DAEMON_OK, Adx::AdxDumpRecord::Instance().UnInit()); | ||
| 152 | + | ||
| 153 | + // PrepareFork + PostForkChild: child detaches the ghost thread, releases the inherited queue, | ||
| 154 | + // clears the flag, and unlocks the mutex -> clean restartable state | ||
| 155 | + Adx::AdxDumpRecord::Instance().Init(""); | ||
| 156 | + EXPECT_EQ(IDE_DAEMON_OK, Adx::AdxDumpRecord::Instance().StartRecord()); | ||
| 157 | + Adx::AdxDumpRecord::PrepareFork(); | ||
| 158 | + Adx::AdxDumpRecord::PostForkChild(); | ||
| 159 | + EXPECT_EQ(false, Adx::AdxDumpRecord::Instance().dumpRecordFlag_); | ||
| 160 | + EXPECT_EQ(nullptr, Adx::AdxDumpRecord::Instance().hostDumpDataInfoQueue_.get()); | ||
| 161 | + EXPECT_EQ(false, Adx::AdxDumpRecord::Instance().recordThread_.joinable()); | ||
| 162 | + | ||
| 163 | + // after the child reset, dump can be rebuilt and torn down again without hang | ||
| 164 | + EXPECT_EQ(IDE_DAEMON_OK, Adx::AdxDumpRecord::Instance().Init("")); | ||
| 165 | + EXPECT_EQ(IDE_DAEMON_OK, Adx::AdxDumpRecord::Instance().StartRecord()); | ||
| 166 | + EXPECT_EQ(IDE_DAEMON_OK, Adx::AdxDumpRecord::Instance().UnInit()); | ||
| 167 | +} | ||
| 168 | + | ||
| 169 | +// real fork: child rebuilds its own dump pipeline, parent keeps running | ||
| 170 | +TEST_F(ADX_DUMP_RECORD_TEST, ForkChildRebuildDump) | ||
| 171 | +{ | ||
| 172 | + Adx::AdxDumpRecord::Instance().Init(""); | ||
| 173 | + EXPECT_EQ(IDE_DAEMON_OK, Adx::AdxDumpRecord::Instance().StartRecord()); | ||
| 174 | + | ||
| 175 | + pid_t pid = fork(); | ||
| 176 | + if (pid == 0) { | ||
| 177 | + // child: pthread_atfork(PostForkChild) already reset state; rebuild a fresh pipeline | ||
| 178 | + int32_t initRet = Adx::AdxDumpRecord::Instance().Init(""); | ||
| 179 | + int32_t startRet = Adx::AdxDumpRecord::Instance().StartRecord(); | ||
| 180 | + int32_t uninitRet = Adx::AdxDumpRecord::Instance().UnInit(); | ||
| 181 | + _exit((initRet == IDE_DAEMON_OK && startRet == IDE_DAEMON_OK && | ||
| 182 | + uninitRet == IDE_DAEMON_OK) ? 0 : 1); | ||
| 183 | + } else if (pid > 0) { | ||
| 184 | + int status = 0; | ||
| 185 | + waitpid(pid, &status, 0); | ||
| 186 | + EXPECT_TRUE(WIFEXITED(status)); | ||
| 187 | + EXPECT_EQ(0, WEXITSTATUS(status)); | ||
| 188 | + // parent's own pipeline is intact and can be torn down cleanly | ||
| 189 | + EXPECT_EQ(IDE_DAEMON_OK, Adx::AdxDumpRecord::Instance().UnInit()); | ||
| 190 | + } | ||
| 191 | +} | ||
| 192 | + | ||
| 88 | TEST_F(ADX_DUMP_RECORD_TEST, UpdateDumpInitNum) | 193 | TEST_F(ADX_DUMP_RECORD_TEST, UpdateDumpInitNum) |
| 89 | { | 194 | { |
| 90 | Adx::AdxDumpRecord::Instance().UpdateDumpInitNum(true); | 195 | Adx::AdxDumpRecord::Instance().UpdateDumpInitNum(true); |
| @@ -154,6 +259,8 @@ TEST_F(ADX_DUMP_RECORD_TEST, RecordDumpDataToQueue) | |||
| 154 | MsgProto *msg = Adx::AdxMsgProto::CreateMsgPacket(IDE_DUMP_REQ, 0, nullptr, dataLen); | 259 | MsgProto *msg = Adx::AdxMsgProto::CreateMsgPacket(IDE_DUMP_REQ, 0, nullptr, dataLen); |
| 155 | Adx::SharedPtr<MsgProto> msgPtr(msg, IdeXfree); | 260 | Adx::SharedPtr<MsgProto> msgPtr(msg, IdeXfree); |
| 156 | Adx::HostDumpDataInfo info = {msgPtr, dataLen}; | 261 | Adx::HostDumpDataInfo info = {msgPtr, dataLen}; |
| 262 | + Adx::AdxDumpRecord::Instance().Init(""); // 复位队列 quit_，使 Push 可入队 | ||
| 263 | + Adx::AdxDumpRecord::Instance().dumpRecordFlag_ = false; // 让 RecordDumpInfo 排空后即退出，避免同步调用阻塞 | ||
| 157 | bool ret = Adx::AdxDumpRecord::Instance().RecordDumpDataToQueue(info); | 264 | bool ret = Adx::AdxDumpRecord::Instance().RecordDumpDataToQueue(info); |
| 158 | EXPECT_EQ(true, ret); | 265 | EXPECT_EQ(true, ret); |
| 159 | Adx::AdxDumpRecord::Instance().RecordDumpInfo(); | 266 | Adx::AdxDumpRecord::Instance().RecordDumpInfo(); |
| @@ -228,6 +335,33 @@ TEST_F(ADX_DUMP_RECORD_TEST, RecordDumpDataToFullQueueLimit) | |||
| 228 | EXPECT_EQ(false, ret); | 335 | EXPECT_EQ(false, ret); |
| 229 | } | 336 | } |
| 230 | 337 | ||
| 338 | +TEST_F(ADX_DUMP_RECORD_TEST, PushAfterQuitRejected) | ||
| 339 | +{ | ||
| 340 | + const char *srcFile = "adx_data_dump_server_manager"; | ||
| 341 | + uint32_t dataLen = strlen(srcFile) + 1 + sizeof(Adx::DumpChunk); | ||
| 342 | + MsgProto *msg = Adx::AdxMsgProto::CreateMsgPacket(IDE_DUMP_REQ, 0, nullptr, dataLen); | ||
| 343 | + Adx::SharedPtr<MsgProto> msgPtr(msg, IdeXfree); | ||
| 344 | + Adx::HostDumpDataInfo info = {msgPtr, dataLen}; | ||
| 345 | + | ||
| 346 | + MOCKER(sysinfo).stubs().will(invoke(SysinfoAmpleMem)); | ||
| 347 | + BoundQueueMemory<HostDumpDataInfo> mem; | ||
| 348 | + | ||
| 349 | + // Quit 之前正常入队 | ||
| 350 | + EXPECT_EQ(true, mem.Push(info)); | ||
| 351 | + EXPECT_EQ(1u, mem.Size()); | ||
| 352 | + | ||
| 353 | + // Quit 之后拒绝入队，队列大小不再增长（避免 join 后数据滞留丢失） | ||
| 354 | + mem.Quit(); | ||
| 355 | + EXPECT_EQ(false, mem.Push(info)); | ||
| 356 | + EXPECT_EQ(1u, mem.Size()); | ||
| 357 | + | ||
| 358 | + // Init 复位：清空上一轮残留数据 + 恢复入队能力 | ||
| 359 | + mem.Init(); | ||
| 360 | + EXPECT_EQ(0u, mem.Size()); // 残留数据被清空 | ||
| 361 | + EXPECT_EQ(true, mem.Push(info)); | ||
| 362 | + EXPECT_EQ(1u, mem.Size()); | ||
| 363 | +} | ||
| 364 | + | ||
| 231 | TEST_F(ADX_DUMP_RECORD_TEST, RecordDumpInfoToMindspore) | 365 | TEST_F(ADX_DUMP_RECORD_TEST, RecordDumpInfoToMindspore) |
| 232 | { | 366 | { |
| 233 | const char *srcFile = "adx_data_dump_server_manager"; | 367 | const char *srcFile = "adx_data_dump_server_manager"; |
| @@ -236,6 +370,8 @@ TEST_F(ADX_DUMP_RECORD_TEST, RecordDumpInfoToMindspore) | |||
| 236 | MsgProto *msg = Adx::AdxMsgProto::CreateMsgPacket(IDE_DUMP_REQ, 0, nullptr, dataLen); | 370 | MsgProto *msg = Adx::AdxMsgProto::CreateMsgPacket(IDE_DUMP_REQ, 0, nullptr, dataLen); |
| 237 | Adx::SharedPtr<MsgProto> msgPtr(msg, IdeXfree); | 371 | Adx::SharedPtr<MsgProto> msgPtr(msg, IdeXfree); |
| 238 | Adx::HostDumpDataInfo info = {msgPtr, dataLen}; | 372 | Adx::HostDumpDataInfo info = {msgPtr, dataLen}; |
| 373 | + Adx::AdxDumpRecord::Instance().Init(""); // 复位队列 quit_，使 Push 可入队 | ||
| 374 | + Adx::AdxDumpRecord::Instance().dumpRecordFlag_ = false; // 让 RecordDumpInfo 排空后即退出，避免同步调用阻塞 | ||
| 239 | bool ret = Adx::AdxDumpRecord::Instance().RecordDumpDataToQueue(info); | 375 | bool ret = Adx::AdxDumpRecord::Instance().RecordDumpDataToQueue(info); |
| 240 | EXPECT_EQ(true, ret); | 376 | EXPECT_EQ(true, ret); |
| 241 | Adx::AdxDumpRecord::Instance().RecordDumpInfo(); | 377 | Adx::AdxDumpRecord::Instance().RecordDumpInfo(); |
| @@ -249,6 +385,8 @@ TEST_F(ADX_DUMP_RECORD_TEST, RecordOptimizedMode) | |||
| 249 | MsgProto *msg = Adx::AdxMsgProto::CreateMsgPacket(IDE_DUMP_REQ, 0, nullptr, dataLen); | 385 | MsgProto *msg = Adx::AdxMsgProto::CreateMsgPacket(IDE_DUMP_REQ, 0, nullptr, dataLen); |
| 250 | Adx::SharedPtr<MsgProto> msgPtr(msg, IdeXfree); | 386 | Adx::SharedPtr<MsgProto> msgPtr(msg, IdeXfree); |
| 251 | Adx::HostDumpDataInfo info = {msgPtr, dataLen}; | 387 | Adx::HostDumpDataInfo info = {msgPtr, dataLen}; |
| 388 | + Adx::AdxDumpRecord::Instance().Init(""); // 复位队列 quit_，使 Push 可入队 | ||
| 389 | + Adx::AdxDumpRecord::Instance().dumpRecordFlag_ = false; // 让 RecordDumpInfo 排空后即退出，避免同步调用阻塞 | ||
| 252 | bool ret = Adx::AdxDumpRecord::Instance().RecordDumpDataToQueue(info); | 390 | bool ret = Adx::AdxDumpRecord::Instance().RecordDumpDataToQueue(info); |
| 253 | EXPECT_EQ(true, ret); | 391 | EXPECT_EQ(true, ret); |
| 254 | uint64_t statsItem = DUMP_STATS_MAX | DUMP_STATS_MIN | DUMP_STATS_AVG | DUMP_STATS_NAN | DUMP_STATS_NEG_INF | DUMP_STATS_POS_INF; | 392 | uint64_t statsItem = DUMP_STATS_MAX | DUMP_STATS_MIN | DUMP_STATS_AVG | DUMP_STATS_NAN | DUMP_STATS_NEG_INF | DUMP_STATS_POS_INF; |
| @@ -86,7 +86,7 @@ TEST_F(ADX_DATADUMP_SERVER_UTEST, AdxDataDumpServerInit_AdxDumpRecord_Failed) | |||
| 86 | TEST_F(ADX_DATADUMP_SERVER_UTEST, AdxDataDumpServerInit_CreateRecordProcess_Failed) | 86 | TEST_F(ADX_DATADUMP_SERVER_UTEST, AdxDataDumpServerInit_CreateRecordProcess_Failed) |
| 87 | { | 87 | { |
| 88 | MOCKER(rtGetRunMode).stubs().will(returnValue(1)); | 88 | MOCKER(rtGetRunMode).stubs().will(returnValue(1)); |
| 89 | - MOCKER_CPP(&Thread::CreateDetachTaskWithDefaultAttr).stubs().will(returnValue(EN_ERROR)); | 89 | + MOCKER_CPP(&Adx::AdxDumpRecord::StartRecord).stubs().will(returnValue(IDE_DAEMON_ERROR)); |
| 90 | EXPECT_EQ(IDE_DAEMON_ERROR, AdxDataDumpServerInit()); | 90 | EXPECT_EQ(IDE_DAEMON_ERROR, AdxDataDumpServerInit()); |
| 91 | EXPECT_EQ(Adx::AdxDumpRecord::Instance().GetDumpInitNum(), 0); | 91 | EXPECT_EQ(Adx::AdxDumpRecord::Instance().GetDumpInitNum(), 0); |
| 92 | EXPECT_EQ(IDE_DAEMON_OK, AdxDataDumpServerUnInit()); | 92 | EXPECT_EQ(IDE_DAEMON_OK, AdxDataDumpServerUnInit()); |
| @@ -98,8 +98,7 @@ TEST_F(ADX_DATADUMP_SERVER_UTEST, AdxDataDumpServerInit_CreateServerProcess_Fail | |||
| 98 | MOCKER(rtGetRunMode).stubs().will(returnValue(1)); | 98 | MOCKER(rtGetRunMode).stubs().will(returnValue(1)); |
| 99 | MOCKER_CPP(&Thread::CreateDetachTaskWithDefaultAttr) | 99 | MOCKER_CPP(&Thread::CreateDetachTaskWithDefaultAttr) |
| 100 | .stubs() | 100 | .stubs() |
| 101 | - .will(returnValue(EN_OK)) | 101 | + .will(returnValue(EN_ERROR)); |
| 102 | - .then(returnValue(EN_ERROR)); | ||
| 103 | EXPECT_EQ(IDE_DAEMON_ERROR, AdxDataDumpServerInit()); | 102 | EXPECT_EQ(IDE_DAEMON_ERROR, AdxDataDumpServerInit()); |
| 104 | EXPECT_EQ(Adx::AdxDumpRecord::Instance().GetDumpInitNum(), 0); | 103 | EXPECT_EQ(Adx::AdxDumpRecord::Instance().GetDumpInitNum(), 0); |
| 105 | EXPECT_EQ(IDE_DAEMON_OK, AdxDataDumpServerUnInit()); | 104 | EXPECT_EQ(IDE_DAEMON_OK, AdxDataDumpServerUnInit()); |
🟡 Medium Priority
在
RecordDumpInfo()(第 725 行) 和RecordDumpDataToQueue()(第 775 行) 中新增的日志调用hostDumpDataInfoQueue_->Size()读取std::queue::size()，但未持有队列内部的mtx_互斥锁。BoundQueueMemory::Size()的实现直接返回dataQueue_.size()，不加锁： uint32_t Size() const { return dataQueue_.size(); }而
Push()/Pop()在持有mtx_的情况下并发修改dataQueue_。在Pop()返回（释放锁）之后立即调用Size()，或在Push()返回后立即调用Size()，与另一个线程的Push/Pop操作形成对dataQueue_容器的无锁并发读写——这是 C++ 标准定义的数据竞争（UB）。实际影响有限（日志值可能略微不准），但从正确性角度属于未定义行为。