Inconsistent order of enqueue and try_dequeue

Two threads：
one thread continuously generates data per 10ms about and enqueue();
the other thread continuously reads data and processes it. If the queue is empty, it will sleep for 20ms,

but the data obtained by try_dequeue is messy

the code like this:

// thread 1
void SendMessage(message) {
  // message 1:  id = 1
  // message 2:  id = 2
  // message 2:  id = 3
  message_queue_.enqueue(message);
}

// thread 2
void ReceiveMessage() {
   while (is_running_ || message_queue_.size_approx() > 0) {
    if (message_queue_.size_approx() <= 0) {
      std::this_thread::sleep_for(std::chrono::milliseconds(20));
      continue;
    }
    CHECK(message_queue_.try_dequeue(message));
    // message 1: id = 3
    // message 2: id = 2
    //  message 3: id = 1
  }
}

So what is the reason for this phenomenon?

Start with the enqueue, try_dequeue, and size_approx calls in the provided two-thread example, then read the queue's ordering and concurrency semantics. Reproduce the scenario with the stated timing and determine whether the observed order is expected or indicates a defect; document the reason and any required correction.