webrtc 的TaskQueue() 任务队列

TaskQueue 定义

见文件：rtc_base\task_queue.h

具体实现

class RTC_LOCKABLE RTC_EXPORT TaskQueue {
 public:
  // TaskQueue priority levels. On some platforms these will map to thread
  // priorities, on others such as Mac and iOS, GCD queue priorities.
  using Priority = ::webrtc::TaskQueueFactory::Priority;

 // 注意这个构造函数，以TaskQueueBase智能指针作为参数
 // TaskQueue 的真正实现，其实是这个TaskQueueBase 
  explicit TaskQueue(std::unique_ptr<webrtc::TaskQueueBase,
                                     webrtc::TaskQueueDeleter> task_queue);
  ~TaskQueue();
  
}

创建一个 TaskQueue

  //首先声明相应对象：一个任务队列工厂、一个任务队列对象
  std::shared_ptr<webrtc::TaskQueueFactory> video_encoder_task_queue_factory_;
  rtc::TaskQueue video_encoder_queue_;
  
  //在指定类构造函数列表中创建任务队列
  //假如我们的调用类是VideoHandler
  //具体的实现如下：
  VideoHandbler::VideoHandbler()
  :  video_encoder_task_queue_factory_(webrtc::CreateDefaultTaskQueueFactory()),
     video_encoder_queue_(video_encoder_task_queue_factory_->CreateTaskQueue(
              "VideoEncoderQueue",
              TaskQueueFactory::Priority::NORMAL)){
     }
//具体过程：
//首先创建一个默认的任务队列工厂
//然后基于任务队列工厂创建一个任务队列TaskQueueBase，
//这个新创建的任务队列工厂TaskQueueBase，作为构造函数参数传给了 TaskQueue()，从而创建了我们需要的video_encoder_queue_


//TaskQueue的构造函数，以TaskQueueBase的智能指针作为参数
//通过成员变量impl_ 接收该指针
//后面会发现，任务最终都传给了TaskQueueBase()
 TaskQueue::TaskQueue(
    std::unique_ptr<webrtc::TaskQueueBase, webrtc::TaskQueueDeleter> task_queue)
    : impl_(task_queue.release()) {}

//向TaskQueue投递任务时，最终还是通过 imple_抛给了TaskQueueBase()
void TaskQueue::PostTask(std::unique_ptr<webrtc::QueuedTask> task) {
  return impl_->PostTask(std::move(task));
}

TaskQueueLibevent

//每个平台都有各自的实现
//我们主要看TaskQueueLibevent()的实现

//TaskQueueLibevent 继承自TaskQueueBase()，真正处理Task的函数
class TaskQueueLibevent final : public TaskQueueBase {
public:
  TaskQueueLibevent(absl::string_view queue_name, rtc::ThreadPriority priority);

  void Delete() override;
  void PostTask(std::unique_ptr<QueuedTask> task) override;
  void PostDelayedTask(std::unique_ptr<QueuedTask> task,
                       uint32_t milliseconds) override;
private:
  bool is_active_ = true;
  //输入、输出管道用来唤起线程
  int wakeup_pipe_in_ = -1;
  int wakeup_pipe_out_ = -1;
  event_base* event_base_;
  event wakeup_event_;
  //任务队列线程
  //一个任务队列对象一个线程
  //对于windows平台内部会执行 CreateThread()创建线程
  //android调用 pthread_create()创建线程
  rtc::PlatformThread thread_;
  Mutex pending_lock_;
  absl::InlinedVector<std::unique_ptr<QueuedTask>, 4> pending_
      RTC_GUARDED_BY(pending_lock_);
  // Holds a list of events pending timers for cleanup when the loop exits.
  std::list<TimerEvent*> pending_timers_;
  
  
  //具体的实现部分
  
  TaskQueueLibevent::TaskQueueLibevent(absl::string_view queue_name,
                                     rtc::ThreadPriority priority)
    : event_base_(event_base_new()),
      thread_(&TaskQueueLibevent::ThreadMain, this, queue_name, priority) {
  int fds[2];
  //创建一个管道用于线程之间通信
  RTC_CHECK(pipe(fds) == 0);
  //把管道设置诶非阻塞的
  SetNonBlocking(fds[0]);
  SetNonBlocking(fds[1]);
  wakeup_pipe_out_ = fds[0];
  wakeup_pipe_in_ = fds[1];
 //绑定管道可读事件，当管道可读时会自动调用 OnWakeup
 //这块是借用libevent库来实现的
  EventAssign(&wakeup_event_, event_base_, wakeup_pipe_out_,
              EV_READ | EV_PERSIST, OnWakeup, this);
  event_add(&wakeup_event_, 0);
  thread_.Start();
}
//调用 fcntl把管道设置诶非阻塞
bool SetNonBlocking(int fd) {
  const int flags = fcntl(fd, F_GETFL);
  RTC_CHECK(flags != -1);
  return (flags & O_NONBLOCK) || fcntl(fd, F_SETFL, flags | O_NONBLOCK) != -1;
}

//线程函数一直轮训相关事件
void TaskQueueLibevent::ThreadMain(void* context) {
  TaskQueueLibevent* me = static_cast<TaskQueueLibevent*>(context);

  {
    CurrentTaskQueueSetter set_current(me);
    while (me->is_active_)
      event_base_loop(me->event_base_, 0);
  }

  for (TimerEvent* timer : me->pending_timers_)
    delete timer;
}

向任务队列中抛入任务

void TaskQueueLibevent::PostTask(std::unique_ptr<QueuedTask> task) {
  {
  //向队列投递任务时，先加锁
    MutexLock lock(&pending_lock_);
    bool had_pending_tasks = !pending_.empty();
    //把任务插入到队列中
    pending_.push_back(std::move(task));

    // Only write to the pipe if there were no pending tasks before this one
    // since the thread could be sleeping. If there were already pending tasks
    // then we know there's either a pending write in the pipe or the thread has
    // not yet processed the pending tasks. In either case, the thread will
    // eventually wake up and process all pending tasks including this one.
    //如果当前线程有未处理的阻塞任务，就暂时不唤醒
    if (had_pending_tasks) {
      return;
    }
  }

  // Note: This behvior outlined above ensures we never fill up the pipe write
  // buffer since there will only ever be 1 byte pending.
  //写入的内容是KRunTasks，表明是要处理的Task任务
  char message = kRunTasks;
  //向管道中写入数据，唤起线程
  RTC_CHECK_EQ(write(wakeup_pipe_in_, &message, sizeof(message)),
               sizeof(message));
}

管道中写入数据后，会触发OnWakeup()函数进行处理

void TaskQueueLibevent::OnWakeup(int socket,
                                 short flags,  // NOLINT
                                 void* context) {
  TaskQueueLibevent* me = static_cast<TaskQueueLibevent*>(context);
  RTC_DCHECK(me->wakeup_pipe_out_ == socket);
  char buf;
  // 调用 read 函数从管道中读入数据
  //读到数据后基于相应的数据类型进行处理
  RTC_CHECK(sizeof(buf) == read(socket, &buf, sizeof(buf)));
  switch (buf) {
    case kQuit:
      me->is_active_ = false;
      event_base_loopbreak(me->event_base_);
      break;
    case kRunTasks: {//要处理的任务
      absl::InlinedVector<std::unique_ptr<QueuedTask>, 4> tasks;
      {
       //加锁
       //并取出要处理的任务
        MutexLock lock(&me->pending_lock_);
        tasks.swap(me->pending_);
      }
      RTC_DCHECK(!tasks.empty());
      //遍历所有的task，并调用各自的Run()函数进行处理
      //最终就是回调传进来任务函数
      for (auto& task : tasks) {
        if (task->Run()) {
          task.reset();
        } else {
          // |false| means the task should *not* be deleted.
          task.release();
        }
      }
      break;
    }
    default:
      RTC_NOTREACHED();
      break;
  }
}

使用例子，向任务队列投入任务

向队列中投入一个lambda 任务,
当任务线程接收到该任务后，就会回调该lambda函数

 video_encoder_queue_.PostTask([this,&video_data]{
       ....
       encoder(video_data);
       ....
       
  });

视频编码队列 encoder_queue_

encoder_queue_ 就是利用前面讲到TaskQueue进行视频编码的

具体实现：

//声明编码队列
rtc::TaskQueue encoder_queue_;
//构造函数列表中创建该编码队列
VideoStreamEncoder::VideoStreamEncoder():
      //创建编码队列
      encoder_queue_(task_queue_factory->CreateTaskQueue(
          "EncoderQueue",
          TaskQueueFactory::Priority::NORMAL))
          {}
          
//向编码队列中投入视频数据进行编码
//所以真正的编码线程，就是encoder_queue所拥有的线程了
void VideoStreamEncoder::OnFrame(const VideoFrame& video_frame) {
      encoder_queue_.PostTask(
      [this, incoming_frame, post_time_us, log_stats, post_interval_us]() {
         if (posted_frames_waiting_for_encode == 1 && !cwnd_frame_drop) {
          MaybeEncodeVideoFrame(incoming_frame, post_time_us);
        }
      }
}
          

析构函数会主动关闭任务队列

析构函数中会自动关闭该任务队列，并停止对应任务线程，不需要我们干预

TaskQueue::~TaskQueue() {
  // There might running task that tries to rescheduler itself to the TaskQueue
  // and not yet aware TaskQueue destructor is called.
  // Calling back to TaskQueue::PostTask need impl_ pointer still be valid, so
  // do not invalidate impl_ pointer until Delete returns.
  impl_->Delete();
}

void TaskQueueLibevent::Delete() {
  RTC_DCHECK(!IsCurrent());
  struct timespec ts;
  char message = kQuit;
  //向管道中写入关闭消息
  while (write(wakeup_pipe_in_, &message, sizeof(message)) != sizeof(message)) {
    // The queue is full, so we have no choice but to wait and retry.
    RTC_CHECK_EQ(EAGAIN, errno);
    ts.tv_sec = 0;
    ts.tv_nsec = 1000000;
    nanosleep(&ts, nullptr);
  }
  //停止线程
  thread_.Stop();
  
  event_del(&wakeup_event_);

  IgnoreSigPipeSignalOnCurrentThread();
  //关闭管道
  close(wakeup_pipe_in_);
  close(wakeup_pipe_out_);
  wakeup_pipe_in_ = -1;
  wakeup_pipe_out_ = -1;

  event_base_free(event_base_);
  delete this;
}

以上就是webrtc的 TaskQueue() 任务队列的实现过程了