Introduction

ROS 2 introduces the concept of managed nodes, also called LifecycleNodes. In the following tutorial, we explain the purpose of these nodes, what makes them different from regular nodes and how they comply to a lifecycle management. Managed nodes contain a state machine with a set of predefined states. These states can be changed by invoking a transition id which indicates the succeeding consecutive state. The state machine is implemented as described at the ROS 2 design page.

Our implementation differentiates between Primary States and Transition States. Primary States are supposed to be steady states in which any node can do the respected task. On the other hand, Transition States are meant as temporary intermediate states attached to a transition. The result of these intermediate states are used to indicate whether a transition between two primary states is considered successful or not. Thus, any managed node can be in one of the following states:

Primary States (steady states):

• unconfigured
• inactive
• active
• shutdown

Transition States (intermediate states):

• configuring
• activating
• deactivating
• cleaningup
• shuttingdown

The possible transitions to invoke are:

• configure
• activate
• deactivate
• cleanup
• shutdown

For a more verbose explanation on the applied state machine, we refer to the design page which provides an in-detail explanation about each state and transition.

The demo

What\'s happening

The demo is split into 3 separate applications:

• lifecycle_talker
• lifecycle_listener
• lifecycle_service_client

The lifecycle_talker represents a managed node and publishes according to which state the node is in. We split the tasks of the talker node into separate pieces and execute them as follows:

1. configuring: We initialize our publisher and timer
2. activate: We activate the publisher and timer in order to enable a publishing
3. deactivate: We stop the publisher and timer
4. cleanup: We destroy the publisher and timer

This demo shows a typical talker/listener pair of nodes. However, imagine a real scenario with attached hardware which may have a rather long booting phase, i.e. a laser or camera. One could imagine bringing up the device driver in the configuring state, start and stop only the publishing of the device\'s data in active/deactive state, and only in the cleanup/shutdown state actually shutdown the device.

The lifecycle_listener is a simple listener which shows the characteristics of the lifecycle talker. The talker enables message publishing only in the active state and thus the listener only receives messages when the talker is in an active state.

The lifecycle_service_client is a script calling different transitions on the lifecycle_talker. This is meant as the external user controlling the lifecycle of nodes.

Run the demo

In order to run this demo, we open three terminals and source our ROS 2 environment variables either from the binary distributions or the workspace we compiled from source.

lifecycle_talker lifecycle_listener lifecycle_service_client

$ ros2 run lifecycle lifecycle_talker $ ros2 run lifecycle lifecycle_listener $ ros2 run lifecycle lifecycle_service_client

Alternatively, these three programs can be run together in the same terminal using the launch file:

ros2 launch lifecycle lifecycle_demo_launch.py

If we look at the output of the lifecycle_talker, we notice that nothing seems to happen. This makes sense, since every node starts as unconfigured. The lifecycle_talker is not configured yet and in our example, no publishers and timers are created yet. The same behavior can be seen for the lifecycle_listener, which is less surprising given that no publishers are available at this moment. The interesting part starts with the third terminal. In there we launch our lifecycle_service_client which is responsible for changing the states of the lifecycle_talker.

Triggering transition 1 (configure)

[lc_client] Transition 1 successfully triggered.
[lc_client] Node lc_talker has current state inactive.

Makes the lifecycle talker change its state to inactive. Inactive means that all publishers and timers are created and configured. However, the node is still not active. Therefore no messages are getting published.

[lc_talker] on_configure() is called.
Lifecycle publisher is currently inactive. Messages are not published.
...

At the same time the lifecycle listener receives a notification as it listens to every state change notification of the lifecycle talker. In fact, the listener receives two consecutive notifications. One for changing from the primary state \"unconfigured\" to \"configuring\", and a second notification changing the state from \"configuring\" to \"inactive\" (since the configuring step was successful in the talker).

[lc_listener] notify callback: Transition from state unconfigured to configuring
[lc_listener] notify callback: Transition from state configuring to inactive

Triggering transition 2 (activate)

[lc_client] Transition 2 successfully triggered.
[lc_client] Node lc_talker has current state active.

Makes the lifecycle talker change its state to active. That means all publishers and timers are now activated and therefore the messages are now getting published.

[lc_talker] on_activate() is called.
[lc_talker] Lifecycle publisher is active. Publishing: [Lifecycle HelloWorld #11]
[lc_talker] Lifecycle publisher is active. Publishing: [Lifecycle HelloWorld #12]
...

The lifecycle listener receives the same set of notifications as before. Lifecycle talker changed its state from inactive to active.

[lc_listener]: notify callback: Transition from state inactive to activating
[lc_listener]: notify callback: Transition from state activating to active

The difference from the earlier transition event is that our listener now also receives the actual published data.

[lc_listener] data_callback: Lifecycle HelloWorld #11
[lc_listener] data_callback: Lifecycle HelloWorld #12
...

Please note that the index of the published message is already at 11. The purpose of this demo is to show that even though we call publish at every state of the lifecycle talker, the messages are only actually published when the state in active.

For the rest of the demo, you will see similar output as we deactivate and activate the lifecycle talker and finally shut it down.

The demo code

lifecycle_talker, lifecycle_listener and lifecycle_service_client

If we have a look at the code, there is one significant change for the lifecycle talker compared to a regular talker. Our node does not inherit from the regular rclcpp::node::Node but from rclcpp_lifecycle::LifecycleNode.

class LifecycleTalker : public rclcpp_lifecycle::LifecycleNode

Every child of LifecycleNodes have a set of callbacks provided. These callbacks go along with the applied state machine attached to it. These callbacks are:

rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn
on_configure(const rclcpp_lifecycle::State & previous_state)

rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn
on_activate(const rclcpp_lifecycle::State & previous_state)

rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn
on_deactivate(const rclcpp_lifecycle::State & previous_state)

rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn
on_cleanup(const rclcpp_lifecycle::State & previous_state)

rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface::CallbackReturn
on_shutdown(const rclcpp_lifecycle::State & previous_state)

In the following we assume that we are inside the namespace rclcpp_lifecycle::node_interfaces::LifecycleNodeInterface to shorten the name of the return type. All these callbacks have a positive default return value (return CallbackReturn::SUCCESS). This allows a lifecycle node to change its state even though no explicit callback function was overridden. There is one other callback function for error handling. Whenever a state transition throws an uncaught exception, we call on_error:

• CallbackReturn on_error(const rclcpp_lifecycle::State & previous_state)

This gives room for executing custom error handling. Only (!) in the case that this function returns CallbackReturn::SUCCESS, the state machine transitions to the state unconfigured. By default, the on_error returns CallbackReturn::FAILURE and the state machine transitions into finalized.

At the same time, every lifecycle node has by default 5 different communication interfaces.

• Publisher <node_name>__transition_event: publishes in case a transition is happening.

This allows users to get notified of transition events within the network.

• Service <node_name>__get_state: query about the current state of the node.

Return either a primary or transition state.

• Service <node_name>__change_state: triggers a transition for the current node.

This service call takes a transition id. The transition is fulfilled only in the case that this transition ID is a valid transition from the current state. All other cases are ignored.

• Service <node_name>__get_available_states: This is meant to be an introspection tool.

It returns a list of all possible states this node can be.

• Service <node_name>__get_available_transitions: Same as above, meant to an introspection tool.

It returns a list of all possible transitions this node can execute.

ros2 lifecycle command line interface

The lifecycle_service_client application is a fixed order script for demo purposes only. It explains the use and the API calls made for this lifecycle implementation, but may be inconvenient to use otherwise. For this reason we implemented a command line tool which lets you dynamically change states or various nodes.

In the case you want to get the current state of the lc_talker node, you would call:

$ ros2 lifecycle get /lc_talker
unconfigured [1]

The next step would be to execute a state change:

$ ros2 lifecycle set /lc_talker configure
Transitioning successful

In order to see what states are currently available:

$ ros2 lifecycle list lc_talker
- configure [1]
  Start: unconfigured
  Goal: configuring
- shutdown [5]
  Start: unconfigured
  Goal: shuttingdown

In this case we see that currently, the available transitions are configure and shutdown. The complete state machine can be viewed with the following command, which can be helpful for debugging or visualization purposes:

$ ros2 lifecycle list lc_talker -a
- configure [1]
  Start: unconfigured
  Goal: configuring
- transition_success [10]
  Start: configuring
  Goal: inactive
- transition_failure [11]
  Start: configuring
  Goal: unconfigured
- transition_error [12]
  Start: configuring
  Goal: errorprocessing

[...]

- transition_error [62]
  Start: errorprocessing
  Goal: finalized

All of the above commands are nothing more than calling the lifecycle node\'s services. With that being said, we can also call these services directly with the ros2 command line interface:

$ ros2 service call /lc_talker/get_state lifecycle_msgs/GetState
requester: making request: lifecycle_msgs.srv.GetState_Request()

response:
lifecycle_msgs.srv.GetState_Response(current_state=lifecycle_msgs.msg.State(id=1, label='unconfigured'))

In order to trigger a transition, we call the change_state service

$ ros2 service call /lc_talker/change_state lifecycle_msgs/ChangeState "{transition: {id: 2}}"
requester: making request: lifecycle_msgs.srv.ChangeState_Request(transition=lifecycle_msgs.msg.Transition(id=2, label=''))

response:
lifecycle_msgs.srv.ChangeState_Response(success=True)

It is slightly less convenient, because you have to know the IDs which correspond to each transition. You can find them though in the lifecycle_msgs package.

$ ros2 interface show lifecycle_msgs/msg/Transition

Changelog for package lifecycle

0.33.2 (2024-03-28)

• A few uncrustify fixes for 0.78. (#667)
• Update maintainer list in package.xml files (#665)
• Contributors: Chris Lalancette, Michael Jeronimo

0.33.1 (2024-02-07)

0.33.0 (2024-01-24)

• Migrate std::bind calls to lambda expressions (#659)
• Contributors: Felipe Gomes de Melo

0.32.1 (2023-12-26)

0.32.0 (2023-11-06)

0.31.1 (2023-09-07)

0.31.0 (2023-08-21)

• Switch to using RCLCPP logging macros in the lifecycle package. (#644)
• Contributors: Chris Lalancette

0.30.1 (2023-07-11)

0.30.0 (2023-06-12)

0.29.0 (2023-06-07)

0.28.1 (2023-05-11)

0.28.0 (2023-04-27)

0.27.0 (2023-04-13)

0.26.0 (2023-04-11)

• update launch file name format to match documentation (#588)
• Contributors: Patrick Wspanialy

0.25.0 (2023-03-01)

0.24.1 (2023-02-24)

0.24.0 (2023-02-14)

• Update the demos to C++17. (#594)
• [rolling] Update maintainers - 2022-11-07 (#589)
• Contributors: Audrow Nash, Chris Lalancette

0.23.0 (2022-11-02)

0.22.0 (2022-09-13)

0.21.0 (2022-04-29)

0.20.1 (2022-04-08)

• Make lifecycle demo automatically exit when done (#558)
• Contributors: Shane Loretz

0.20.0 (2022-03-01)

• Use default on_activate()/on_deactivate() implemenetation of Node (#552)
• Contributors: Ivan Santiago Paunovic

0.19.0 (2022-01-14)

0.18.0 (2021-12-17)

• Update maintainers to Audrow Nash and Michael Jeronimo (#543)
• Contributors: Audrow Nash

0.17.0 (2021-10-18)

• Fix use of future in lifecycle demo (#534)
• Fixing deprecated subscriber callback warnings (#532)
• Contributors: Abrar Rahman Protyasha, Christophe Bedard

0.16.0 (2021-08-11)

0.15.0 (2021-05-14)

0.14.2 (2021-04-26)

• Cleanup the README.rst for the lifecycle demo. (#508)
• Contributors: Chris Lalancette

0.14.1 (2021-04-19)

0.14.0 (2021-04-06)

• change ParameterEventHandler to take events as const ref instead of shared pointer (#494)
• Contributors: William Woodall

0.13.0 (2021-03-25)

0.12.1 (2021-03-18)

0.12.0 (2021-01-25)

0.11.0 (2020-12-10)

• Update the package.xml files with the latest Open Robotics maintainers (#466)
• Contributors: Michael Jeronimo

0.10.1 (2020-09-21)

• Add missing required parameter in LifecycleNode launch action (#456)
• Contributors: Ivan Santiago Paunovic

0.10.0 (2020-06-17)

0.9.3 (2020-06-01)

0.9.2 (2020-05-26)

• Fix typo (#445)
• Replace ros2 msg command in lifecycle README (#446)
• Contributors: Audrow Nash, Shota Aoki

0.9.1 (2020-05-12)

0.9.0 (2020-04-30)

• Replace deprecated launch_ros usage (#437)
• Update launch_ros action usage (#431)
• code style only: wrap after open parenthesis if not in one line (#429)
• Contributors: Dirk Thomas, Jacob Perron

0.8.4 (2019-11-19)

0.8.3 (2019-11-11)

0.8.2 (2019-11-08)

• Remove unnecessary dependency on ros2run (#413)
• Contributors: Michel Hidalgo

0.8.1 (2019-10-23)

• Replace ready_fn with ReadyToTest action (#404)
• Contributors: Peter Baughman

0.8.0 (2019-09-26)

• Fix lifecycle_service_client namespace (#369)
• Contributors: Cameron Evans

0.7.6 (2019-05-30)

0.7.5 (2019-05-29)

• Update asciinema recordings (#360)
• Use rate instead of thread::sleep to react to Ctrl-C (#348)
• Contributors: Dirk Thomas, Karsten Knese

0.7.4 (2019-05-20)

• Add lifecycle rostest (#336)
• Contributors: Michel Hidalgo

0.7.3 (2019-05-10)

0.7.2 (2019-05-08)

• changes to avoid deprecated API\'s (#332)
• Corrected publish calls with shared_ptr signature (#327)
• Contributors: William Woodall, ivanpauno

0.7.1 (2019-04-26)

0.7.0 (2019-04-14)

• Updated for NodeOptions Node constructor. (#308)
• Contributors: Michael Carroll

0.6.2 (2019-01-15)

• Added readme.rst (#300)
• Contributors: Karsten Knese

0.6.1 (2018-12-13)

0.6.0 (2018-12-07)

• Cleaned up lifecycle demo (#283)
• Updated for refactoring in rclcpp (#276)
• Added semicolons to all RCLCPP and RCUTILS macros. (#278)
• Fixed typo in comment (#270)
• Contributors: Chris Lalancette, Karsten Knese, Yutaka Kondo

0.5.1 (2018-06-28)

0.5.0 (2018-06-27)

• Converted launch files to the new launch style. (#262)
• Updated to support remapping arguments to python nodes by passing unused arguments to rclpy from argparse. (#252)
• Updated to handle change in signature to get_service_name. (#245)
• Updated launch files to account for the \"old launch\" getting renamespaced as launch -> launch.legacy. (#239)
• Updated service client demos to handle multiple requests. (#228)
• Contributors: Geoffrey Biggs, Kevin Allen, Shane Loretz, William Woodall, dhood