You're reading the documentation for a development version. For the latest released version, please have a look at Foxy.

Working with multiple ROS 2 middleware implementations

This page explains the default RMW implementation and how to specify an alternative.

Specifying RMW implementations

To have multiple RMW implementations available for use you must have installed our binaries and any additional dependencies for specific RMW implementations, or built ROS 2 from source with multiple RMW implementations in the workspace (they are included by default and their dependencies are met). See Install DDS implementations.


Starting in Beta 2 and above both C++ and Python nodes support an environment variable RMW_IMPLEMENTATION. To choose a different RMW implemenation you can set the environment variable RMW_IMPLEMENTATION to a specific implementation identifier.

To run the talker demo using the C++ and listener using Python with the RMW implementation for Connext:

RMW_IMPLEMENTATION=rmw_connextdds ros2 run demo_nodes_cpp talker

# Run in another terminal
RMW_IMPLEMENTATION=rmw_connextdds ros2 run demo_nodes_py listener

Adding RMW implementations to your workspace

Suppose that you have built your ROS 2 workspace with only Cyclone DDS installed and therefore only the Cyclone DDS RMW implementation built. The last time your workspace was built, any other RMW implementation packages, rmw_connextdds for example, were probably unable to find installations of the relevant DDS implementations. If you then install an additional DDS implementation, Connext for example, you will need to re-trigger the check for a Connext installation that occurs when the Connext RMW implementation is being built. You can do this by specifying the --cmake-force-configure flag on your next workspace build, and you should see that the RMW implementation package then gets built for the newly installed DDS implementation.

It is possible to run into a problem when “rebuilding” the workspace with an additional RMW implementation using the --cmake-force-configure option where the build complains about the default RMW implementation changing. To resolve this, you can either set the default implementation to what is was before with the RMW_IMPLEMENTATION CMake argument or you can delete the build folder for packages that complain and continue the build with --start-with <package name>.

Troubleshooting

Ensuring use of a particular RMW implementation

If the RMW_IMPLEMENTATION environment variable is set to an RMW implementation for which support is not installed, you will see an error message similar to the following if you have only one implementation installed:

Expected RMW implementation identifier of 'rmw_connextdds' but instead found 'rmw_fastrtps_cpp', exiting with 102.

If you have support for multiple RMW implementations installed and you request use of one that is not installed, you will see something similar to:

Error getting RMW implementation identifier / RMW implementation not installed (expected identifier of 'rmw_connextdds'), exiting with 1.

If this occurs, double check that your ROS 2 installation includes support for the RMW implementation that you have specified in the RMW_IMPLEMENTATION environment variable.

If you want to switch between RMW implementations, verify that the ROS 2 daemon process is not running with the previous RMW implementation to avoid any issues between nodes and command line tools such as ros2 node. For example, if you run:

RMW_IMPLEMENTATION=rmw_connextdds ros2 run demo_nodes_cpp talker

and

ros2 node list

it will generate a daemon with a Cyclone DDS implementation:

21318 22.0  0.6 535896 55044 pts/8    Sl   16:14   0:00 /usr/bin/python3 /opt/ros/rolling/bin/_ros2_daemon --rmw-implementation rmw_cyclonedds_cpp --ros-domain-id 22

Even if you run the command line tool again with the correct RMW implementation, the daemon’s RMW implementation will not change and the ROS 2 command line tools will fail.

To solve this, simply stop the daemon process:

ros2 daemon stop

and rerun the ROS 2 command line tool with the correct RMW implementation.

RTI Connext on OSX: Failure due to insufficient shared memory kernel settings

If you receive an error message similar to below when running RTI Connext on OSX:

[D0062|ENABLE]DDS_DomainParticipantPresentation_reserve_participant_index_entryports:!enable reserve participant index
[D0062|ENABLE]DDS_DomainParticipant_reserve_participant_index_entryports:Unusable shared memory transport. For a more in-   depth explanation of the possible problem and solution, please visit https://community.rti.com/kb/osx510.

This error is caused by an insufficient number or size of shared memory segments allowed by the operating system. As a result, the DomainParticipant is unable to allocate enough resources and calculate its participant index which causes the error.

You can increase the shared memory resources of your machine either temporarily or permanently.

To increase the settings temporarily, you can run the following commands as user root:

/usr/sbin/sysctl -w kern.sysv.shmmax=419430400
/usr/sbin/sysctl -w kern.sysv.shmmin=1
/usr/sbin/sysctl -w kern.sysv.shmmni=128
/usr/sbin/sysctl -w kern.sysv.shmseg=1024
/usr/sbin/sysctl -w kern.sysv.shmall=262144

To increase the settings permanently, you will need to edit or create the file /etc/sysctl.conf. Creating or editing this file will require root permissions. Either add to your existing etc/sysctl.conf file or create /etc/sysctl.conf with the following lines:

kern.sysv.shmmax=419430400
kern.sysv.shmmin=1
kern.sysv.shmmni=128
kern.sysv.shmseg=1024
kern.sysv.shmall=262144

You will need to reboot the machine after modifying this file to have the changes take effect.

This solution is edited from the RTI Connext community forum. See the original post for more detailed explanation.