16.2. Use ROS 2 with Fast-DDS Discovery Server

This section explains how to run some ROS 2 examples using the Discovery Servers as discovery communication. In order to get more information about the specific use of this configuration, please check the Discovery Server Documentation or read the common use cases for this configuration.

The following tutorial gathers the steps to check this functionality and learn how to use it with ROS 2.

The Simple Discovery Protocol is the standard protocol defined in the DDS standard. However, it has certain known disadvantages in some scenarios, mainly:

• It does not Scale efficiently, as the number of exchanged packets highly increases as new nodes are added.

• It requires Multicasting capabilities that may not work reliably in some scenarios, e.g. WiFi.

The Discovery Server provides a Client-Server Architecture that allows the nodes to connect with each other using an intermediate server. Each node will work as a Client, sharing its info with the Discovery Server and receiving the discovery information from it. This means that the network traffic is highly reduced in big systems, and it does not require Multicasting.

These Discovery Servers can be independent, duplicated or connected with each other in order to create redundancy over the network and avoid having a Single-Point-Of-Failure.

16.2.1. Discovery Server v2

The new version v2 of Discovery Server, available from Fast DDS v2.0.2, implements a new filter feature that allows to further reduce the number of discovery messages sent. This version uses the topic of the different nodes to decide if two nodes must be connected, or they could be left unmatched. The following schema represents the decrease of the discovery packages:

This architecture reduces the number of packages sent between the server and the different clients dramatically. In the following graph, the reduction in traffic network over the discovery phase for a RMF Clinic demo use case, is shown:

In order to use this functionality, Fast-DDS Discovery Server can be set using the XML configuration for Participants. Furthermore, Fast DDS provides an easier way to set a Discovery Server communication using the fastdds CLI tool and an environment variable, which are going to be used along this tutorial. For a more detailed explanation about the configuration of the Discovery Server, visit Discovery Server Settings.

16.2.2. Prerequisites

This tutorial assumes you have a working Foxy ROS 2 installation. In case your installation is using a Fast DDS version lower than v2.0.2 you could not use the fastdds tool. You could update your repository to use a different Fast DDS version, or set the discovery server by Fast-DDS XML QoS configuration.

Note

This tutorial can also be run in Galactic exporting the environment variable that selects Fast DDS as the middleware layer:

export RMW_IMPLEMENTATION=rmw_fastrtps_cpp

16.2.3. Run the demo

The talker-listener ROS 2 demo allows to create a talker node that publishes a Hello World message every second, and a listener node that listens to these messages.

By Sourcing ROS 2 you will get access to the CLI of Fast DDS: fastdds. This CLI gives access to the discovery tool, which allows to launch a server. This server will manage the discovery process for the nodes that connect to it.

Important

Do not forget to source ROS 2 in every new terminal opened.

16.2.3.1. Setup Discovery Server

Start by launching a server with id 0, with port 11811 and listening on all available interfaces.

Open a new terminal and run:

fastdds discovery -i 0

16.2.3.2. Launch node listener

Execute the listener demo, that will listen in /chatter topic.

In a new terminal, set the environment variable ROS_DISCOVERY_SERVER to use Discovery Server. (Do not forget to source ROS 2 in every new terminal)

export ROS_DISCOVERY_SERVER="127.0.0.1:11811"

Afterwards, launch the listener node. Use the argument --remap __node:=listener_discovery_server to change the node’s name for future purpose.

ros2 run demo_nodes_cpp listener --ros-args --remap __node:=listener_discovery_server

This process will create a ROS 2 node, that will automatically create a client for the Discovery Server and use the server created previously to run the discovery protocol.

16.2.3.3. Launch node talker

Open a new terminal and set the environment variable as before, so the node raises a client for the discovery protocol.

export ROS_DISCOVERY_SERVER="127.0.0.1:11811"
ros2 run demo_nodes_cpp talker --ros-args --remap __node:=talker_discovery_server

Now, we should see the talker publishing Hello World messages, and the listener receiving these messages.

16.2.3.4. Demonstrate Discovery Server execution

So far, there is not proof that this example and the standard talker-listener example run differently. For this purpose, run another node that is not connected to our Discovery Server. Just run a new listener (listening in /chatter topic by default) in a new terminal and check that it is not connected to the talker already running.

ros2 run demo_nodes_cpp listener --ros-args --remap __node:=simple_listener

In this case, we should not see the listener receiving the messages.

To finally verify that everything is running correctly, a new talker can be created using the simple discovery protocol.

ros2 run demo_nodes_cpp talker --ros-args --remap __node:=simple_talker

Now we should see the listener simple_listener receiving the messages from simple_talker but not the other messages from talker_discovery_server.

16.2.4. Advance user cases

The following paragraphs are going to show different features of the Discovery Server that allows to hold a robust structure over the node’s network.

16.2.4.1. Server Redundancy

By using the Fast DDS tool, several servers can be created, and the nodes can be connected to as many servers as desired. This allows to have a safe redundancy network that will work even if some servers or nodes shut down unexpectedly. Next schema shows a simple architecture that will work with server redundancy:

In different terminals, run the next code to establish a communication over redundant servers.

fastdds discovery -i 0 -l 127.0.0.1 -p 11811

fastdds discovery -i 1 -l 127.0.0.1 -p 11888

-i N means server with id N. When referencing the servers with ROS_DISCOVERY_SERVER, server 0 must be in first place and server 1 in second place.

export ROS_DISCOVERY_SERVER="127.0.0.1:11811;127.0.0.1:11888"
ros2 run demo_nodes_cpp talker --ros-args --remap __node:=talker

export ROS_DISCOVERY_SERVER="127.0.0.1:11811;127.0.0.1:11888"
ros2 run demo_nodes_cpp listener --ros-args --remap __node:=listener

Now, if one of these servers fails, there would still be discovery communication between nodes.

16.2.4.2. Backup Server

Fast DDS Discovery Server allows to easily build a server with a backup functionality. This allows the server to retake the last state it saved in case of a shutdown.

In different terminals, run the next code to establish a communication over a backup server.

fastdds discovery -i 0 -l 127.0.0.1 -p 11811 -b

export ROS_DISCOVERY_SERVER="127.0.0.1:11811"
ros2 run demo_nodes_cpp talker --ros-args --remap __node:=talker

export ROS_DISCOVERY_SERVER="127.0.0.1:11811"
ros2 run demo_nodes_cpp listener --ros-args --remap __node:=listener

Several backup files are created in the path the server has run. Two SQLite files and two json files that contains the information required to raise a new server in case of failure, avoiding the whole discovery process to happen again and without losing information.

16.2.4.3. Discovery partitions

The Discovery Server communication could be used with different servers to split in virtual partitions the discovery info. This means that two endpoints only would know each other if there is a server or a server network between them. We are going to execute an example with two different independent servers. The following image shows a schema of the architecture desired:

With this schema Listener 1 will be connected to Talker 1 and Talker 2, as they share Server 1. Listener 2 will connect with Talker 1 as they share Server 2. But Listener 2 will not hear the messages from Talker 2 because they do not share any server or servers’ network that connect them.

Run the first server listening in localhost in default port 11811.

fastdds discovery -i 0 -l 127.0.0.1 -p 11811

In another terminal run the second server listening in localhost in port another port, in this case 11888.

fastdds discovery -i 1 -l 127.0.0.1 -p 11888

Now, run each node in a different terminal. Use the environment variable ROS_DISCOVERY_SERVER to decide which server they are connected to. Be aware that the ids must match (Environment variables).

export ROS_DISCOVERY_SERVER="127.0.0.1:11811;127.0.0.1:11888"
ros2 run demo_nodes_cpp talker --ros-args --remap __node:=talker_1

export ROS_DISCOVERY_SERVER="127.0.0.1:11811;127.0.0.1:11888"
ros2 run demo_nodes_cpp listener --ros-args --remap __node:=listener_1

export ROS_DISCOVERY_SERVER="127.0.0.1:11811"
ros2 run demo_nodes_cpp talker --ros-args --remap __node:=talker_2

export ROS_DISCOVERY_SERVER=";127.0.0.1:11888"
ros2 run demo_nodes_cpp listener --ros-args --remap __node:=listener_2

We should see how Listener 1 is receiving double messages, while Listener 2 is in a different partition from Talker 2 and so it does not listen to it.

Note

Once two endpoints know each other, they do not need the server network between them to listen to each other messages.

16.2.5. ROS 2 Introspection

ROS 2 Command Line Interface (CLI) implements several introspection features to analyze the behaviour of a ROS 2 execution. These features (i.e. rosbag, topic list, etc.) are very helpful to understand a ROS 2 working network.

Most of these features use the DDS capability to share any topic information with every exiting participant. However, the new Discovery Server v2 implements a traffic network reduction that limits the discovery data between nodes that do not share a topic. This means that not every node will receive every topic data unless it has a reader in that topic. As most of ROS 2 CLI Introspection is executed by adding a node into the network (some of them use ROS 2 Daemon, and some create their own nodes), using Discovery Server v2 we will find that most of these functionalities are limited and do not have all the information.

The Discovery Server v2 functionality allows every node running as a SUPER_CLIENT, a kind of Client that connects to a SERVER, from which it receives all the available discovery information (instead of just what it needs). In this sense, ROS 2 introspection tools can be configured as Super Client, thus being able to discover every entity that is using the Discovery Server protocol within the network.

16.2.5.1. Daemon’s related commands

The ROS 2 Daemon is used in several ROS 2 CLI introspection commands. It adds a ROS 2 Node to the network in order to receive all the data sent. In order for the ROS 2 CLI to work when using Discover Server discovery mechanism, the ROS 2 Daemon needs to be configured as Super Client. Therefore, this section is devoted to explain how to use ROS 2 CLI with ROS 2 Daemon running as a Super Client. This will allow the Daemon to discover the entire Node graph, and to receive every topic and endpoint information. To do so, a Fast DDS XML configuration file is used to configure the ROS 2 Daemon and CLI tools.

Warning

Although it is possible to run the ROS 2 Daemon as a Server, this is not recommended since the daemon will stop after two hours of inactivity, taking the Server down with it.

Below you can find a XML configuration file which will configure every new participant as a Super Client.

• XML Super Client configuration file

First of all, instantiate a Discovery Server using Fast DDS CLI

fastdds discovery -i 0 -l 127.0.0.1 -p 11811

Run a talker and a listener that will discover each other through the Server (notice that ROS_DISCOVERY_SERVER configuration is the same as the one in super_client_configuration_file.xml).

export ROS_DISCOVERY_SERVER="127.0.0.1:11811"
ros2 run demo_nodes_cpp listener --ros-args --remap __node:=listener

export ROS_DISCOVERY_SERVER="127.0.0.1:11811"
ros2 run demo_nodes_cpp talker --ros-args --remap __node:=talker

Then, instantiate a ROS 2 Daemon using the Super Client configuration (remember to source ROS 2 installation in every new terminal).

export FASTRTPS_DEFAULT_PROFILES_FILE=super_client_configuration_file.xml
ros2 daemon stop
ros2 daemon start
ros2 topic list
ros2 node info /talker
ros2 topic info /chatter
ros2 topic echo /chatter

We can also see the Node’s Graph using the ROS 2 tool rqt_graph as follows (you may need to press the refresh button):

export FASTRTPS_DEFAULT_PROFILES_FILE=super_client_configuration_file.xml
rqt_graph

16.2.5.2. No Daemon commands

Some ROS 2 CLI tools can be executed without the ROS 2 Daemon. In order for these tools to connect with a Discovery Server and receive all the topics information they need to be instantiated as a Super Client that connects to the Server.

Following the previous configuration, build a simple system with a talker and a listener. First, run a Server:

fastdds discovery -i 0 -l 127.0.0.1 -p 11811

Then, run the talker and listener is separate terminals:

export ROS_DISCOVERY_SERVER="127.0.0.1:11811"
ros2 run demo_nodes_cpp listener --ros-args --remap __node:=listener

export ROS_DISCOVERY_SERVER="127.0.0.1:11811"
ros2 run demo_nodes_cpp talker --ros-args --remap __node:=talker

Continue using the ROS 2 CLI with --no-daemon option with the new configuration. New nodes will connect with the existing Server and will know every topic. Exporting ROS_DISCOVERY_SERVER is not needed as the remote server has been configured in the xml file.

export FASTRTPS_DEFAULT_PROFILES_FILE=super_client_configuration_file.xml
ros2 topic list --no-daemon
ros2 node info /talker --no-daemon --spin-time 2

16.2.6. Compare Discovery Server with Simple Discovery

In order to compare the ROS 2 execution using Simple Discovery or Discovery Server, two scripts that execute a talker and many listeners and analyze the network traffic during this time are provided. For this experiment, tshark is required to be installed on your system. The configuration file is mandatory in order to avoid using intra-process mode.

Note

These scripts require a Discovery Server closure feature that is only available from Fast DDS v2.1.0 and forward. In order to use this functionality, compile ROS 2 with Fast DDS v2.1.0 or higher.

These scripts’ functionalities are references for advance purpose and their study is left to the user.

• bash network traffic generator

• python3 graph generator

• XML configuration

Run the bash script with the setup path to source ROS 2 as argument. This will generate the traffic trace for simple discovery. Executing the same script with second argument SERVER, it will generates the trace for service discovery.

Note

Depending on your configuration of tcpdump, this script may require sudo privileges to read traffic across your network device.

After both executions are done, run the python script to generates a graph similar to the one below:

$ export FASTRTPS_DEFAULT_PROFILES_FILE="no_intraprocess_configuration.xml"
$ sudo bash generate_discovery_packages.bash ~/ros2_foxy/install/local_setup.bash
$ sudo bash generate_discovery_packages.bash ~/ros2_foxy/install/local_setup.bash SERVER
$ python3 discovery_packets.py

This graph is the result of a is a specific example, the user can execute the scripts and watch their own results. It can easily be seen how the network traffic is reduced when using Discovery Service.

The reduction in traffic is a result of avoiding every node announcing itself and waiting a response from every other node in the net. This creates a huge amount of traffic in large architectures. This reduction from this method increases with the number of Nodes, making this architecture more scalable than the simple one.

Since Fast DDS v2.0.2 the new Discovery Server v2 is available, substituting the old Discovery Server. In this new version, those nodes that do not share topics will not know each other, saving the whole discovery data required to connect them and their endpoints. Notice that this is not this example case, but even though the massive reduction could be appreciate due to the hidden architecture topics of ROS 2 nodes.