Rosbag C++ API: how to avoid tremendous consumption of RAM

edit

Hi all,

I wrote a simple class to read sensor_msgs::PointCloud2 messages from a rosbag follwing the cookbook:

#include <rosbag/bag.h>
#include <rosbag/view.h>
#include <sensor_msgs/PointCloud2.h>

template <class M>
class MessageCollector
{
public:
  MessageCollector(const std::string& filename)
  {
    bag_.open(filename, rosbag::bagmode::Read);
  }

  void loadBag(const std::vector<std::string>& topics)
  {

    // create bag view
    rosbag::View view(bag_, rosbag::TopicQuery(topics));

    // Load all messages into the dataset
    BOOST_FOREACH (rosbag::MessageInstance const m, view)
    {
      boost::shared_ptr<M const> msg = m.instantiate<M>();
      if (msg != NULL)
      {
        msgCallback(msg);
      }
    }

    // close bag
    bag_.close();
  }

  void msgCallback(const boost::shared_ptr<M const>& msg)
  {
    dataset_.push_back(msg);
  }

  rosbag::Bag bag_;
  std::vector<boost::shared_ptr<M const>> dataset_;
};

int main()
{

  ros::Time::init();
  MessageCollector<sensor_msgs::PointCloud2> collector(
      "data.bag");
  std::vector<std::string> topics{ "point_cloud_1",
                                   "point_cloud_2",
                                   "point_cloud_3" };
  collector.loadBag(topics);

  return 0;
}

My problem is that the data.bag is ~50GB and contains ~50K lidars messages.

So, after few seconds this executable consumes all of my 32GB of RAM.

Please, can someone explain what is going on and/or what I'm doing wrong?

Thanks.