故障导向安全的架构设计的应用实例-EasyDarwin的日志设计

但是在今天上午发生了一件事情，还是让我很惊奇，这件事情就是RTP会话被超时结束了，但是设备（拉流）还是能建立RTSP会话，只不过拉取不到任何数据，从抓包来看，设备如果拉取不到数据，就会重新建立RTSP会话，具体可以看抓包日志和流媒体服务器的日志。

当推流端开始推送RTP包后，RTSP会话会调用到RTSPSession::HandleIncomingDataPacket()，在这个函数里调用了fRTPSession->RefreshTimeout()，于是我们可以看到fRTPSession里的fTimeoutTask任务，fTimeoutTask的超时时间是又配置文件里的部分的rtp_timeout字段的值，也就是120秒，如果120秒没有收到任何推送的RTP包，fTimeoutTask会触发超时（详情请见TimeoutTaskThread::Run()函数），也就是会给RTPSession这个任务发送一个Task::kTimeoutEvent事件。

0180120

SInt64 TimeoutTaskThread::Run(){ //ok, check for timeouts now. Go through the whole queue OSMutexLocker locker(&fMutex); SInt64 curTime = OS::Milliseconds(); SInt64 intervalMilli = kIntervalSeconds * 1000;//always default to 60 seconds but adjust to smallest interval > 0 SInt64 taskInterval = intervalMilli; for (OSQueueIter iter(&fQueue); !iter.IsDone(); iter.Next()) { TimeoutTask* theTimeoutTask = (TimeoutTask*)iter.GetCurrent()->GetEnclosingObject();  //if it's time to time this task out, signal it if ((theTimeoutTask->fTimeoutAtThisTime > 0) && (curTime >= theTimeoutTask->fTimeoutAtThisTime)) {#if TIMEOUT_DEBUGGING qtss_printf("TimeoutTask %"_S32BITARG_" timed out. Curtime = %"_64BITARG_"d, timeout time = %"_64BITARG_"d\n",(SInt32)theTimeoutTask, curTime, theTimeoutTask->fTimeoutAtThisTime);#endif theTimeoutTask->fTask->Signal(Task::kTimeoutEvent); } else { taskInterval = theTimeoutTask->fTimeoutAtThisTime - curTime; if ( (taskInterval > 0) && (theTimeoutTask->fTimeoutInMilSecs > 0) && (intervalMilli > taskInterval) ) intervalMilli = taskInterval + 1000; // set timeout to 1 second past this task's timeout#if TIMEOUT_DEBUGGING qtss_printf("TimeoutTask %"_S32BITARG_" not being timed out. Curtime = %"_64BITARG_"d. timeout time = %"_64BITARG_"d\n", (SInt32)theTimeoutTask, curTime, theTimeoutTask->fTimeoutAtThisTime);#endif } } (void)this->GetEvents();//we must clear the event mask!  OSThread::ThreadYield(); #if TIMEOUT_DEBUGGING qtss_printf ("TimeoutTaskThread::Run interval seconds= %"_S32BITARG_"\n", (SInt32) intervalMilli/1000);#endif  return intervalMilli;//don't delete me!}

RTPSession收到kTimeoutEvent事件后会向其它模块发送QTSS_ClientSessionClosing_Role角色的事件，至此RTPSession会话任务结束掉。

SInt64 RTPSession::Run(){ //if we have been instructed to go away, then let's delete ourselves if ((events & Task::kKillEvent) || (events & Task::kTimeoutEvent) || (fModuleDoingAsyncStuff)) { if (!fModuleDoingAsyncStuff) { if (events & Task::kTimeoutEvent) fClosingReason = qtssCliSesCloseTimeout;  //deletion is a bit complicated. For one thing, it must happen from within //the Run function to ensure that we aren't getting events when we are deleting //ourselves. We also need to make sure that we aren't getting RTSP requests //(or, more accurately, that the stream object isn't being used by any other //threads). We do this by first removing the session from the session map. #if RTPSESSION_DEBUGGING qtss_printf("RTPSession %"_S32BITARG_": about to be killed. Eventmask = %"_S32BITARG_"\n",(SInt32)this, (SInt32)events);#endif // We cannot block waiting to UnRegister, because we have to // give the RTSPSessionTask a chance to release the RTPSession. OSRefTable* sessionTable = QTSServerInterface::GetServer()->GetRTPSessionMap(); Assert(sessionTable != NULL); if (!sessionTable->TryUnRegister(&fRTPMapElem)) { this->Signal(Task::kKillEvent);// So that we get back to this place in the code return kCantGetMutexIdleTime; }  // The ClientSessionClosing role is allowed to do async stuff fModuleState.curTask = this; fModuleDoingAsyncStuff = true; // So that we know to jump back to the fCurrentModule = 0; // right place in the code  // Set the reason parameter  theParams.clientSessionClosingParams.inReason = fClosingReason;  // If RTCP packets are being generated internally for this stream,  // Send a BYE now. RTPStream** theStream = NULL; UInt32 theLen = 0;  if (this->GetPlayFlags() & qtssPlayFlagsSendRTCP) { SInt64 byePacketTime = OS::Milliseconds(); for (int x = 0; this->GetValuePtr(qtssCliSesStreamObjects, x, (void**)&theStream, &theLen) == QTSS_NoErr; x++) if (theStream && *theStream != NULL) (*theStream)->SendRTCPSR(byePacketTime, true); } }  //at this point, we know no one is using this session, so invoke the //session cleanup role. We don't need to grab the session mutex before //invoking modules here, because the session is unregistered and //therefore there's no way another thread could get involved anyway UInt32 numModules = QTSServerInterface::GetNumModulesInRole(QTSSModule::kClientSessionClosingRole); { for (; fCurrentModule < numModules; fCurrentModule++) {  fModuleState.eventRequested = false; fModuleState.idleTime = 0; QTSSModule* theModule = QTSServerInterface::GetModule(QTSSModule::kClientSessionClosingRole, fCurrentModule); (void)theModule->CallDispatch(QTSS_ClientSessionClosing_Role, &theParams); // If this module has requested an event, return and wait for the event to transpire if (fModuleState.eventRequested) return fModuleState.idleTime; // If the module has requested idle time... } }  return -1;//doing this will cause the destructor to get called. }}

RTSPSession的超时机制和RTPSession类似，RTSPSession的超时时间是配置文件的real_rtsp_timeout字段值，在kFilteringRequest状态时会去刷新fTimeoutTask的时间，如果超时，也会结束整个状态机的处理部分，同时调用fRTPSession的Teardown函数，结束RTPSession会话。

SInt64 RTSPSession::Run(){ ...  //check for a timeout or a kill. If so, just consider the session dead if ((events & Task::kTimeoutEvent) || (events & Task::kKillEvent)) { fLiveSession = false; } ...// 状态机的部分 //fObjectHolders--  if(!IsLiveSession()&& fObjectHolders > 0){   OSRefTable* theMap = QTSServerInterface::GetServer()->GetRTPSessionMap();  OSRef* theRef = theMap->Resolve(&fLastRTPSessionIDPtr);  if (theRef != NULL){  fRTPSession = (RTPSession*)theRef->GetObject();  if(fRTPSession) fRTPSession->Teardown();  theMap->Release(fRTPSession->GetRef());  fRTPSession = NULL;  }  }  // Make absolutely sure there are no resources being occupied by the session // at this point. this->CleanupRequest(); // Only delete if it is ok to delete! if (fObjectHolders == 0) return -1; // If we are here because of a timeout, but we can't delete because someone // is holding onto a reference to this session, just reschedule the timeout. // // At this point, however, the session is DEAD. return 0;

于是从以上可知，RTP会话超时是在120秒内没有收到任何RTP包就结束这个会话，此时RTSP会话还是存在的，也就是拉流端还是能够来请求拉流，鉴权也能通过，当RTSP会话在超过180秒内没有收到任何数据包则会结束整个RTSPSession。每天会更新论文和视频，还有如果想学习c++知识在晚上8.30免费观看这个直播：https://ke.qq.com/course/131973#tuin=b52b9a80

閱讀更多 IT布丁老師 的文章

關鍵字: 架构设计 return 会话