Using Fast DDS Discovery Server as discovery protocol [community-contributed]

Goal: This tutorial will show how to launch ROS 2 Nodes using the Fast DDS Discovery Server discovery protocol.

Tutorial level: Intermediate

Time: 20 minutes

Background

Starting from ROS 2 Eloquent Elusor, the Fast DDS Discovery Server protocol is a feature that offers a centralized dynamic discovery mechanism, as opposed to the distributed mechanism used in DDS by default. This tutorial explains how to run some ROS 2 examples using the Fast DDS Discovery Server feature as discovery communication.

In order to get more information about the available discovery configuration, please check the following documentation or read the Fast DDS Discovery Server specific documentation.

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

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

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

The Fast DDS Discovery Server provides a Client-Server Architecture that allows nodes to connect with each other using an intermediate server. Each node functions as a discovery client, sharing its info with one or more discovery servers and receiving discovery information from it. This reduces discovery-related network traffic and it does not require multicasting capabilities.

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.

Fast DDS Discovery Server v2

The latest ROS 2 Foxy Fitzroy release (December 2020) included a new version, version 2 of the Fast DDS Discovery Server. This version includes a new filter feature that further reduces the number of discovery messages sent. This version uses the topic of the different nodes to decide if two nodes wish to communicate, or if they can be left unmatched (i.e. not discovering each other). The following figure shows the decrease in discovery messages:

This architecture reduces the number of messages sent between the server and clients dramatically. In the following graph, the reduction in network traffic over the discovery phase for the RMF Clinic demonstration is shown:

In order to use this functionality, the discovery server can be configured using the XML configuration for Participants. It is also possible to configure the discovery server using the fastdds tool and an environment variable, which is the approach used in this tutorial. For a more detailed explanation about the configuration of the discovery server, visit the Fast DDS Discovery Server documentation.

Prerequisites

This tutorial assumes you have a ROS 2 Foxy (or newer) installation. If your installation is using a ROS 2 version lower than Foxy, you cannot use the fastdds tool. Thus, in order to use the Discovery Server, you can update your repository to use a different Fast DDS version, or configure the discovery server using the Fast DDS XML QoS configuration.

Run this tutorial

The talker-listener ROS 2 demo creates 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 tool fastdds. This tool gives access to the discovery tool, which can be used to launch a discovery 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.

Setup Discovery Server

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

Open a new terminal and run:

fastdds discovery --server-id 0

Launch listener node

Execute the listener demo, to listen to the /chatter topic.

In a new terminal, set the environment variable ROS_DISCOVERY_SERVER to the location of the discovery server. (Do not forget to source ROS 2 in every new terminal)

export ROS_DISCOVERY_SERVER=127.0.0.1:11811

Launch the listener node. Use the argument --remap __node:=listener_discovery_server to change the node’s name for this tutorial.

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

This will create a ROS 2 node, that will automatically create a client for the discovery server and connect to the server created previously to perform discovery, rather than using multicast.

Launch talker node

Open a new terminal and set the ROS_DISCOVERY_SERVER environment variable as before so that the node starts a discovery client.

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

You should now see the talker publishing “hello world” messages, and the listener receiving these messages.

Demonstrate Discovery Server execution

So far, there is no evidence that this example and the standard talker-listener example are running differently. To clearly demonstrate this, run another node that is not connected to the discovery server. 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

The new listener node should not be receiving the “hello world” messages.

To finally verify that everything is running correctly, a new talker can be created using the simple discovery protocol (the default DDS distributed discovery mechanism) for discovery.

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

Now you should see the simple_listener node receiving the “hello world” messages from simple_talker but not the other messages from talker_discovery_server.

Visualization tool rqt_graph

The rqt_graph tool can be used to verify the nodes and structure of this example. Remember, in order to use rqt_graph with the discovery server protocol (i.e., to see the listener_discovery_server and talker_discovery_server nodes) the ROS_DISCOVERY_SERVER environment variable must be set before launching it.

Advance use cases

The following sections show different features of the discovery server that allow you to build a robust discovery server over the network.

Server Redundancy

By using fastdds tool, multiple discovery servers can be created. Discovery clients (ROS nodes) can connect to as many servers as desired. This allows to have a redundant network that will work even if some servers or nodes shut down unexpectedly. The figure below shows a simple architecture that provides server redundancy.

In several terminals, run the following code to establish a communication with redundant servers.

fastdds discovery --server-id 0 --ip-address 127.0.0.1 --port 11811

fastdds discovery --server-id 1 --ip-address 127.0.0.1 --port 11888

--server-id 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 will still be discovery capability available and nodes will still discover each other.

Backup Server

The Fast DDS Discovery Server allows creating a server with backup functionality. This allows the server to restore the last state it saved in case of a shutdown.

In different terminals, run the following code to establish a communication with a backed-up server.

fastdds discovery --server-id 0 --ip-address 127.0.0.1 --port 11811 --backup

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 discovery server’s working directory (the directory it was launched in). The two SQLite files and two json files contain the information required to start a new server and restore the failed server’s state in case of failure, avoiding the need for the discovery process to happen again, and without losing information.

Discovery partitions

Communication with discovery servers can be split to create virtual partitions in the discovery information. This means that two endpoints will only know about each other if there is a shared discovery server or a network of discovery servers between them. We are going to execute an example with two independent servers. The following figure shows the architecture.

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 discovery server or discovery servers, including indirectly via connections between redundant discovery servers.

Run the first server listening on localhost with the default port of 11811.

fastdds discovery --server-id 0 --ip-address 127.0.0.1 --port 11811

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

fastdds discovery --server-id 1 --ip-address 127.0.0.1 --port 11888

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

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 messages from both talker nodes, while Listener 2 is in a different partition from Talker 2 and so does not receive messages from it.

Note

Once two endpoints (ROS nodes) have discovered each other, they do not need the discovery server network between them to listen to each other’s messages.

Compare Fast DDS Discovery Server with Simple Discovery Protocol

In order to compare executing nodes using the Simple Discovery Protocol (the default DDS mechanism for distributed discovery) or the 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 intraprocess mode.

Note

These scripts require a discovery server closure feature that is only available from versions newer than the version provided in ROS 2 Foxy. In order to use this functionality, compile ROS 2 with Fast DDS v2.1.0 or higher.

These scripts’ features are references for advanced purposes and their study is left to the user.

• bash network traffic generator

• python3 graph generator

• XML configuration

Run the bash script with the path to setup.bash file to source ROS 2 as an argument. This will generate the traffic trace for simple discovery. Execute the same script with second argument SERVER. It will generate the trace for using the discovery server.

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 generate 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 specific run of the experiment. The reader can execute the scripts and generate their own results for comparison. It can easily be seen that 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 on the network. This creates a huge amount of traffic in large architectures. The reduction from this method increases with the number of nodes, making this architecture more scalable than the Simple Discovery Protocol approach.

The new Fast DDS Discovery Server v2 is available since Fast DDS v2.0.2, replacing the old discovery server. In this new version, those nodes that do not share topics will automatically not discover each other, saving the whole discovery data required to connect them and their endpoints. The experiment above does not show this case, but even so the massive reduction in traffic can be appreciated due to the hidden infrastructure topics of ROS 2 nodes.