avoid lifecycle node transition exception

[fujitatomoya]

addresses #1209

replaces #1317

[fujitatomoya]

with this fix, we can avoid the crash on action server from exception and the action client can receive the response with failure as below.

root@tomoyafujita:~/ros2_ws/colcon_ws# ros2 run lifecycle_py lifecycle_talker 
[WARN] [1721693858.103483007] [rcl_lifecycle]: No transition matching 2 found for current state unconfigured
[ERROR] [1721693858.114465029] [lc_talker]: Unable to start transition 2 from current state (1, 'unconfigured')

root@tomoyafujita:~/ros2_ws/colcon_ws# ros2 service call /lc_talker/change_state lifecycle_msgs/srv/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=False)

[fujitatomoya]

@mjforan @Barry-Xu-2018 @chrisbitter if any of you can do the review or test, that will be really appreciated.

[Barry-Xu-2018]

About __change_state(), my understanding is

• Before calling this function, the state machine should be in a primary state (unconfigured, inactive, active or finalized). After calling this function, it should switch to another primary state.
• The argument transition_id for __change_state() should be one of configure, cleanup, activate, deactivate, unconfigured_shutdown, inactive_shutdown and active_shutdown

So, when returning from __change_state(), we need to ensure that the state machine is in a primary state, not a transition state (such as configuring, cleaningup, shuttingdown, etc). If trigger_transition_by_id/trigger_transition_by_label throw exception, it should call on_error and then switch to unconfigured or finalized state.

Is my understanding correct?

Besides, __change_state() may be executed concurrently (the program modifies the state while an external process modifies the state through change_state service). This function does not have lock protection

[mjforan]

Testing the basic lifecycle_py demo shows the change working properly.
Before:

After:

I had to use the Rolling branch of rclpy because lifecycle_py doesn't work under Jazzy.

I agree with @Barry-Xu-2018; the exception should not be handled the same way for all three instances.
My opinion:

x	exception caused by	how to handle
1	invalid transition request	write to log and return error to caller
2	unable to transition out of transition state	enter ErrorProcessing
3	unable to transition out of ErrorProcessing	raise an exception and crash

[fujitatomoya]

@mjcarroll @sloretz @ahcorde can you take a look when you have time?

[ahcorde]

There are some conflict to resolve

[fujitatomoya]

@ahcorde appreciate your time for the review, thanks! yeah, i will do rebase and resolve the conflict, and get back to you.

[fujitatomoya]

@Barry-Xu-2018 about your comment on #1319 (comment) makes sense. (including locking mechanism) AFAIS, current PR is aligned with rclcpp behavior. how about addressing this comment with another issue? this is gonna be more complicated and takes time to verify.

[fujitatomoya]

Pulls: #1319
Gist: https://gist.githubusercontent.com/fujitatomoya/1d9faa816731c8aee120548004e4bb46/raw/7ea0f998c7e152f14f1d178cf76fa6d108383ba1/ros2.repos
BUILD args: --packages-above-and-dependencies rclpy
TEST args: --packages-above rclpy
ROS Distro: rolling
Job: ci_launcher
ci_launcher ran: https://ci.ros2.org/job/ci_launcher/14891

• Linux
• Linux-aarch64
• Linux-rhel
• Windows

[fujitatomoya]

@Barry-Xu-2018 friendly ping

[Barry-Xu-2018]

LGTM

[fujitatomoya]

• Linux-rhel
• Windows

[fujitatomoya]

https://ci.ros2.org/job/ci_linux-rhel/1789/#showFailuresLink is unrelated.