You're reading the documentation for an older, but still supported, version of ROS 2. For information on the latest version, please have a look at Humble.

Recording a bag from your own node

Goal: Record data from your own node to a bag.

Tutorial level: Advanced

Time: 20 minutes

Background

rosbag2 doesn’t just provide the ros2 bag command line tool. It also provides an API for reading from and writing to a bag from your own source code. This allows you to subscribe to a topic and save the received data to a bag at the same time as performing any other processing of your choice on that data.

Prerequisites

You should have the rosbag2 packages installed as part of your regular ROS 2 setup.

If you’ve installed from Debian packages on Linux, it may be installed by default. If it is not, you can install it using this command.

sudo apt install ros-foxy-rosbag2

This tutorial discusses using ROS 2 bags, including from the terminal. You should have already completed the basic ROS 2 bag tutorial.

Tasks

1 Create a package

Open a new terminal and source your ROS 2 installation so that ros2 commands will work.

Navigate into the dev_ws directory created in a previous tutorial. Navigate into the dev_ws/src directory and create a new package:

ros2 pkg create --build-type ament_cmake bag_recorder_nodes --dependencies rclcpp rosbag2_cpp example_interfaces

Your terminal will return a message verifying the creation of your package bag_recorder_nodes and all its necessary files and folders. The --dependencies argument will automatically add the necessary dependency lines to package.xml and CMakeLists.txt. In this case, the package will use the rosbag2_cpp package as well as the rclcpp package. A dependency on the example_interfaces package is also required for later parts of this tutorial.

1.1 Update package.xml

Because you used the --dependencies option during package creation, you don’t have to manually add dependencies to package.xml or CMakeLists.txt. As always, though, make sure to add the description, maintainer email and name, and license information to package.xml.

<description>C++ bag writing tutorial</description>
<maintainer email="you@email.com">Your Name</maintainer>
<license>Apache License 2.0</license>

2 Write the C++ node

Inside the dev_ws/src/bag_recorder_nodes/src directory, create a new file called simple_bag_recorder.cpp and paste the following code into it.

#include <rclcpp/rclcpp.hpp>
#include <std_msgs/msg/string.hpp>

#include <rosbag2_cpp/typesupport_helpers.hpp>
#include <rosbag2_cpp/writer.hpp>
#include <rosbag2_cpp/writers/sequential_writer.hpp>
#include <rosbag2_storage/serialized_bag_message.hpp>

using std::placeholders::_1;

class SimpleBagRecorder : public rclcpp::Node
{
public:
  SimpleBagRecorder()
  : Node("simple_bag_recorder")
  {
    const rosbag2_cpp::StorageOptions storage_options({"my_bag", "sqlite3"});
    const rosbag2_cpp::ConverterOptions converter_options(
      {rmw_get_serialization_format(),
       rmw_get_serialization_format()});
    writer_ = std::make_unique<rosbag2_cpp::writers::SequentialWriter>();

    writer_->open(storage_options, converter_options);

    writer_->create_topic(
      {"chatter",
       "std_msgs/msg/String",
       rmw_get_serialization_format(),
       ""});

    subscription_ = create_subscription<std_msgs::msg::String>(
      "chatter", 10, std::bind(&SimpleBagRecorder::topic_callback, this, _1));
  }

private:
  void topic_callback(std::shared_ptr<rclcpp::SerializedMessage> msg) const
  {
    auto bag_message = std::make_shared<rosbag2_storage::SerializedBagMessage>();

    bag_message->serialized_data = std::shared_ptr<rcutils_uint8_array_t>(
      new rcutils_uint8_array_t,
      [this](rcutils_uint8_array_t *msg) {
        auto fini_return = rcutils_uint8_array_fini(msg);
        delete msg;
        if (fini_return != RCUTILS_RET_OK) {
          RCLCPP_ERROR(get_logger(),
            "Failed to destroy serialized message %s", rcutils_get_error_string().str);
        }
      });
    *bag_message->serialized_data = msg->release_rcl_serialized_message();

    bag_message->topic_name = "chatter";
    if (rcutils_system_time_now(&bag_message->time_stamp) != RCUTILS_RET_OK) {
      RCLCPP_ERROR(get_logger(), "Error getting current time: %s",
        rcutils_get_error_string().str);
    }

    writer_->write(bag_message);
  }

  rclcpp::Subscription<rclcpp::SerializedMessage>::SharedPtr subscription_;
  std::unique_ptr<rosbag2_cpp::writers::SequentialWriter> writer_;
};

int main(int argc, char * argv[])
{
  rclcpp::init(argc, argv);
  rclcpp::spin(std::make_shared<SimpleBagRecorder>());
  rclcpp::shutdown();
  return 0;
}

2.1 Examine the code

The #include statements at the top are the package dependencies. Note the inclusion of headers from the rosbag2_cpp package for the functions and structures necessary to work with bag files.

In the class constructor we begin by creating the writer object we will use to write to the bag. We must provide the storage options for the bag. These specify the name (my_bag) and format (sqlite3) of the bag. We must also provide conversion options, which specify how data input into the writer will be serialised and how that same data should be serialised when written to the bag. In most cases you can specify these as the same value as no conversion of serialisation formats is neccessary. We use the rmw_get_serialization_format() function to retrieve the serialisation format used by the underlying middleware. This is the serialisation format that data will be received in, and so is the format we store it in the bag as.

Finally, the third line creates the writer object. We create a SequentialWriter, which is the simplest writer object. It does not perform advanced operations like compressing the data as it is written to the bag.

rosbag2_cpp::StorageOptions storage_options({"my_bag", "sqlite3"});
rosbag2_cpp::ConverterOptions converter_options({rmw_get_serialization_format(), rmw_get_serialization_format()});
writer_ = std::make_unique<rosbag2_cpp::writers::SequentialWriter>();

Now that we have a writer object, we can open the bag using it.

writer_->open(storage_options, converter_options);

The next step is to inform the writer of each topic that we will write to the bag. This is done by calling create_topic and passing in an instance of the rosbag2_storage::TopicMetadata structure. Here we are using modern C++ syntax to construct an instance of this structure in place rather than creating it separately and passing it in. The arguments stored in the rosbag2_storage::TopicMetadata structure are:

• The name of the topic. Note that this does not need to be the same as the topic the data is received on.

• The type of data in the topic. This must be the same as the type of data being stored in the bag.

• The serialisation format of the data. As before, we simply use the same serialisation format as the underlying middleware.

• Any QoS settings to specify for the topic. These must be specified in YAML format. Leaving it as an empty string will use the system defaults.

writer_->create_topic({"chatter", "std_msgs/msg/String", rmw_get_serialization_format(), ""});

With the writer now set up to record data we pass to it, we create a subscription and specify a callback for it. We will write data to the bag in the callback.

subscription_ = create_subscription<std_msgs::msg::String>(
  "chatter", 10, std::bind(&SimpleBagRecorder::topic_callback, this, _1));

The callback itself is different from a typical callback. Rather than receiving an instance of the data type of the topic, we instead receive a rclcpp::SerializedMessage. We do this for two reasons.

1. The message data will need to be serialised before being written to the bag, so rather than unserialising it when receiving the data and then re-serialising it, we ask ROS to just give us the serialised message as-is.

2. The writer API requires a serialised message, so by asking for a serialised message from ROS we save ourselves the effort of serialising the data ourselves.

void topic_callback(std::shared_ptr<rclcpp::SerializedMessage> msg) const
{

Within the subscription callback, the first thing to do is to create an instance of rosbag2_storage::SerializedBagMessage. This is the data type that represents a single data sample in a bag.

auto bag_message = std::make_shared<rosbag2_storage::SerializedBagMessage>();

The next block of code is somewhat complex due to the memory management requirements of working directly with serialised data from the middleware. The memory of the serialised data is owned by the SerializedMessage object, but the SerializedBagMessage object we will pass to the bag must own the memory. Otherwise the memory might go out of scope and be deleted before it is written to the bag, causing a memory access error. To prevent this, we call release_rcl_serialized_message() on the SerializedMessage object. This causes it to release its ownership of the memory, allowing the SerializedBagMessage to take ownership.

However we also need to ensure that the SerializedBagMessage object will delete the memory properly when it is cleaned up. This is achieved by providing a custom deleter function when creating the serialized_data member of the bag_message instance. This is the purpose of the lambda function being passed into the constructor of the std::shared_ptr<rcutils_uint8_array_t> object (i.e. the serialized_data member of the bag_message object). When the shared_ptr goes out of scope, the lambda function will be called and will clean up the memory.

bag_message->serialized_data = std::shared_ptr<rcutils_uint8_array_t>(
  new rcutils_uint8_array_t,
  [this](rcutils_uint8_array_t *msg) {
    auto fini_return = rcutils_uint8_array_fini(msg);
    delete msg;
    if (fini_return != RCUTILS_RET_OK) {
      RCLCPP_ERROR(get_logger(),
        "Failed to destroy serialized message %s", rcutils_get_error_string().str);
    }
  });
*bag_message->serialized_data = msg->release_rcl_serialized_message();

The next line is used to tell the writer what topic this sample is for. This is necessary as the serialised message contains no type information and could, by error, be written to any topic in the bag.

bag_message->topic_name = "chatter";

The time stamp of the message must also be set in the time_stamp member of the bag_message objecT. This can be anything appropriate to your data, but two common values are the time at which the data was produced, if known, and the time it is received. The second option, the time of reception, is used here.

if (rcutils_system_time_now(&bag_message->time_stamp) != RCUTILS_RET_OK) {
  RCLCPP_ERROR(get_logger(), "Error getting current time: %s",
    rcutils_get_error_string().str);
}

The final step in the callback is to pass the data to the writer object so it can be written to the bag.

writer_->write(bag_message);

The class contains two member variables.

1. The subscription object. Note that the template parameter is the type of the callback, not the type of the topic. In this case the callback receives a rclcpp::SerializedMessage shared pointer, so this is what the template parameter must be.

2. A managed pointer to the writer object used to write to the bag. Note the type of writer used here is the rosbag2_cpp::writers::SequentialWriter. Other writers may be available with different behaviours.

rclcpp::Subscription<rclcpp::SerializedMessage>::SharedPtr subscription_;
std::unique_ptr<rosbag2_cpp::writers::SequentialWriter> writer_;

The file finishes with the main function used to create an instance of the node and start ROS processing it.

int main(int argc, char * argv[])
{
  rclcpp::init(argc, argv);
  rclcpp::spin(std::make_shared<SimpleBagRecorder>());
  rclcpp::shutdown();
  return 0;
}

2.2 Add executable

Now open the CMakeLists.txt file. Below the dependency find_package(rosbag2_cpp REQUIRED) add the following lines of code.

add_executable(simple_bag_recorder src/simple_bag_recorder.cpp)
ament_target_dependencies(simple_bag_recorder rclcpp rosbag2_cpp)

install(TARGETS
  simple_bag_recorder
  DESTINATION lib/${PROJECT_NAME}
)

3 Build and run

Navigate back to the root of your workspace, dev_ws, and build your new package.

LinuxmacOSWindows

colcon build --packages-select bag_recorder_nodes

Open a new terminal, navigate to dev_ws, and source the setup files.

LinuxmacOSWindows

source install/setup.bash

Create a directory for the bag. This directory will contain all the files that form a single bag.

LinuxmacOSWindows

mkdir my_bag

(If the my_bag directory already exists, you must first delete it before re-creating it.)

Now run the node:

ros2 run bag_recorder_nodes simple_bag_recorder

Open a second terminal and run the talker example node.

ros2 run demo_nodes_cpp talker

This will start publishing data on the chatter topic. As the bag-writing node receives this data, it will write it to the my_bag bag.

Terminate both nodes. Then, in one terminal start the listener example node.

ros2 run demo_nodes_cpp listener

In the other terminal, use ros2 bag to play the bag recorded by your node.

ros2 bag play my_bag

You will see the messages from the bag being received by the listener node.

4 Record synthetic data from a node

Any data can be recorded into a bag, not just data received over a topic. A common use case for writing to a bag from your own node is to generate and store synthetic data. In this section you will learn how to write a node that generates some data and stores it in a bag. We will demonstrate two approaches for doing this. The first uses a node with a timer; this is the approach that you would use if your data generation is external to the node, such as reading data directly from hardware (e.g. a camera). The second approach does not use a node; this is the approach you can use when you do not need to use any functionality from the ROS infrastructure.

4.1 Write a C++ node

Inside the dev_ws/src/bag_recorder_nodes/src directory, create a new file called data_generator_node.cpp and paste the following code into it.

#include <chrono>

#include <example_interfaces/msg/int32.hpp>
#include <rclcpp/rclcpp.hpp>
#include <rclcpp/serialization.hpp>

#include <rosbag2_cpp/writer.hpp>
#include <rosbag2_cpp/writers/sequential_writer.hpp>
#include <rosbag2_storage/serialized_bag_message.hpp>

using namespace std::chrono_literals;

class DataGenerator : public rclcpp::Node
{
public:
  DataGenerator()
  : Node("data_generator")
  {
    data.data = 0;
    const rosbag2_cpp::StorageOptions storage_options({"timed_synthetic_bag", "sqlite3"});
    const rosbag2_cpp::ConverterOptions converter_options(
      {rmw_get_serialization_format(),
       rmw_get_serialization_format()});
    writer_ = std::make_unique<rosbag2_cpp::writers::SequentialWriter>();

    writer_->open(storage_options, converter_options);

    writer_->create_topic(
      {"synthetic",
       "example_interfaces/msg/Int32",
       rmw_get_serialization_format(),
       ""});

    timer_ = create_wall_timer(1s, std::bind(&DataGenerator::timer_callback, this));
  }

private:
  void timer_callback()
  {
    auto serializer = rclcpp::Serialization<example_interfaces::msg::Int32>();
    auto serialized_message = rclcpp::SerializedMessage();
    serializer.serialize_message(&data, &serialized_message);

    auto bag_message = std::make_shared<rosbag2_storage::SerializedBagMessage>();

    bag_message->serialized_data = std::shared_ptr<rcutils_uint8_array_t>(
      new rcutils_uint8_array_t,
      [this](rcutils_uint8_array_t *msg) {
        auto fini_return = rcutils_uint8_array_fini(msg);
        delete msg;
        if (fini_return != RCUTILS_RET_OK) {
          RCLCPP_ERROR(get_logger(),
            "Failed to destroy serialized message %s", rcutils_get_error_string().str);
        }
      });
    *bag_message->serialized_data = serialized_message.release_rcl_serialized_message();

    bag_message->topic_name = "synthetic";
    if (rcutils_system_time_now(&bag_message->time_stamp) != RCUTILS_RET_OK) {
      RCLCPP_ERROR(get_logger(), "Error getting current time: %s",
        rcutils_get_error_string().str);
    }

    writer_->write(bag_message);
    ++data.data;
  }

  rclcpp::TimerBase::SharedPtr timer_;
  std::unique_ptr<rosbag2_cpp::writers::SequentialWriter> writer_;
  example_interfaces::msg::Int32 data;
};

int main(int argc, char * argv[])
{
  rclcpp::init(argc, argv);
  rclcpp::spin(std::make_shared<DataGenerator>());
  rclcpp::shutdown();
  return 0;
}

4.2 Examine the code

Much of this code is the same as the first example. The important differences are described here.

First, the name of the bag is changed.

rosbag2_cpp::StorageOptions storage_options({"timed_synthetic_bag", "sqlite3"});

The topic name and data type that will be stored are also different, so the writer needs to be told this.

writer_->create_topic({"synthetic", "example_interfaces/msg/Int32", rmw_get_serialization_format(), ""});

Rather than a subscription to a topic, this node has a timer. The timer fires with a one-second period, and calls the given member function when it does.

timer_ = create_wall_timer(1s, std::bind(&DataGenerator::timer_callback, this));

Within the timer callback, we generate (or otherwise obtain, e.g. read from a serial port connected to some hardware) the data we wish to store in the bag. The important difference between this and the previous sample is that the data is not yet serialised. Because the bag writer expects serialised data, we must serialise it first. This can be done using the rclcpp::Serialization class.

auto serializer = rclcpp::Serialization<example_interfaces::msg::Int32>();
auto serialized_message = rclcpp::SerializedMessage();
serializer.serialize_message(&data, &serialized_message);

The remainder of the code in the callback is the same, modified slightly to account for topic name and data type differences, and to increment the data each time the callback is executed.

4.3 Add executable

Open the CMakeLists.txt file and add the following lines after the previously-added lines (specifically, after the install(TARGETS ...) macro call).

add_executable(data_generator_node src/data_generator_node.cpp)
ament_target_dependencies(data_generator_node rclcpp rosbag2_cpp example_interfaces)

install(TARGETS
  data_generator_node
  DESTINATION lib/${PROJECT_NAME}
)

4.4 Build and run

Navigate back to the root of your workspace, dev_ws, and build your package.

LinuxmacOSWindows

colcon build --packages-select bag_recorder_nodes

Open a new terminal, navigate to dev_ws, and source the setup files.

LinuxmacOSWindows

source install/setup.bash

Create a directory for the bag. This directory will contain all the files that form a single bag.

LinuxmacOSWindows

mkdir timed_synthetic_bag

(If the timed_synthetic_bag directory already exists, you must first delete it before re-creating it.)

Now run the node:

ros2 run bag_recorder_nodes data_generator_node

Wait for 30 seconds or so, then terminate the node with ctrl-c. Next, play back the created bag.

ros2 bag play timed_synthetic_bag

Open a second terminal and echo the /synthetic topic.

ros2 topic echo /synthetic

You will see the data that was generated and stored in the bag printed to the console at a rate of one message per second.

5 Record synthetic data from an executable

Now that you can create a bag that stores data from a source other than a topic, you will learn how to generate and record synthetic data from a non-node executable. The advantage of this approach is simpler code and rapid creation of a large quantity of data.

5.1 Write a C++ executable

Inside the dev_ws/src/bag_recorder_nodes/src directory, create a new file called data_generator_executable.cpp and paste the following code into it.

#include <iostream>

#include <rclcpp/rclcpp.hpp>
#include <rclcpp/serialization.hpp>
#include <example_interfaces/msg/int32.hpp>

#include <rosbag2_cpp/writer.hpp>
#include <rosbag2_cpp/writers/sequential_writer.hpp>
#include <rosbag2_storage/serialized_bag_message.hpp>

int main(int, char**)
{
  example_interfaces::msg::Int32 data;
  data.data = 0;
  const rosbag2_cpp::StorageOptions storage_options({"big_synthetic_bag", "sqlite3"});
  const rosbag2_cpp::ConverterOptions converter_options(
    {rmw_get_serialization_format(),
     rmw_get_serialization_format()});
  std::unique_ptr<rosbag2_cpp::writers::SequentialWriter> writer_ =
    std::make_unique<rosbag2_cpp::writers::SequentialWriter>();

  writer_->open(storage_options, converter_options);

  writer_->create_topic(
    {"synthetic",
     "example_interfaces/msg/Int32",
     rmw_get_serialization_format(),
     ""});

  rcutils_time_point_value_t time_stamp;
  if (rcutils_system_time_now(&time_stamp) != RCUTILS_RET_OK) {
    std::cerr << "Error getting current time: " <<
      rcutils_get_error_string().str;
    return 1;
  }
  for (int32_t ii = 0; ii < 100; ++ii) {
    auto serializer = rclcpp::Serialization<example_interfaces::msg::Int32>();
    auto serialized_message = rclcpp::SerializedMessage();
    serializer.serialize_message(&data, &serialized_message);

    auto bag_message = std::make_shared<rosbag2_storage::SerializedBagMessage>();

    bag_message->serialized_data = std::shared_ptr<rcutils_uint8_array_t>(
      new rcutils_uint8_array_t,
      [](rcutils_uint8_array_t *msg) {
        auto fini_return = rcutils_uint8_array_fini(msg);
        delete msg;
        if (fini_return != RCUTILS_RET_OK) {
          std::cerr << "Failed to destroy serialized message " <<
            rcutils_get_error_string().str;
        }
      });
    *bag_message->serialized_data = serialized_message.release_rcl_serialized_message();

    bag_message->topic_name = "synthetic";
    bag_message->time_stamp = time_stamp;

    writer_->write(bag_message);
    ++data.data;
    time_stamp += 1000000000;
  }

  return 0;
}

5.2 Examine the code

A comparison of this sample and the previous sample will reveal that they are not that different. The only significant difference is the use of a for loop to drive the data generation rather than a timer.

Notice that we are also now generating time stamps for the data rather than relying on the current system time for each sample. The time stamp can be any value you need it to be. The data will be played back at the rate given by these time stamps, so this is a useful way to control the default playback speed of the samples. Notice also that while the gap between each sample is a full second in time, this executable does not need to wait a second between each sample. This allows us to generate a lot of data covering a wide span of time in much less time than playback will take.

rcutils_time_point_value_t time_stamp;
if (rcutils_system_time_now(&time_stamp) != RCUTILS_RET_OK) {
  std::cerr << "Error getting current time: " <<
    rcutils_get_error_string().str;
  return 1;
}
for (int32_t ii = 0; ii < 100; ++ii) {
  ...
  bag_message->time_stamp = time_stamp;
  ...
  time_stamp += 1000000000;
}

5.3 Add executable

Open the CMakeLists.txt file and add the following lines after the previously-added lines.

add_executable(data_generator_executable src/data_generator_executable.cpp)
ament_target_dependencies(data_generator_executable rclcpp rosbag2_cpp example_interfaces)

install(TARGETS
  data_generator_executable
  DESTINATION lib/${PROJECT_NAME}
)

5.4 Build and run

Navigate back to the root of your workspace, dev_ws, and build your package.

LinuxmacOSWindows

colcon build --packages-select bag_recorder_nodes

Open a terminal, navigate to dev_ws, and source the setup files.

LinuxmacOSWindows

source install/setup.bash

Create a directory for the bag. This directory will contain all the files that form a single bag.

LinuxmacOSWindows

mkdir big_synthetic_bag

(If the big_synthetic_bag directory already exists, you must first delete it before re-creating it.)

Now run the executable:

ros2 run bag_recorder_nodes data_generator_executable

Note that the executable runs and finishes very quickly.

Now play back the created bag.

ros2 bag play big_synthetic_bag

Open a second terminal and echo the /synthetic topic.

ros2 topic echo /synthetic

You will see the data that was generated and stored in the bag printed to the console at a rate of one message per second. Even though the bag was generated rapidly it is still played back at the rate the time stamps indicate.

Summary

You created a node that records data it receives on a topic into a bag. You tested recording a bag using the node, and verified the data was recorded by playing back the bag. You then went on to create a node and an executable to generate synthetic data and store it in a bag.