:farmer: `test_launch_ros` generating build regressions in nightly_win by a windows access violation

[miguelgonrod]

Generated by Generative AI

No response

Operating System:

Windows 11

ROS version or commit hash:

rolling

RMW implementation (if applicable):

No response

RMW Configuration (if applicable):

No response

Client library (if applicable):

No response

'ros2 doctor --report' output

Steps to reproduce issue

1. Go to https://ci.ros2.org/job/nightly_win_rel or https://ci.ros2.org/job/nightly_win_rep
2. Build a Job
3. See the job failing

Expected behavior

To build correctly

Actual behavior

Failing build

Additional information

Reference Build

• nightly_win_rel#3780
• nightly_win_rep#4087

Log Output

--- stderr: test_launch_ros
Windows fatal exception: access violation

Thread 0x00002768 (most recent call first):
  File "C:\ci\ws\install\Lib\site-packages\rclpy\executors.py", line 872 in _wait_for_ready_callbacks
  File "C:\ci\ws\install\Lib\site-packages\rclpy\executors.py", line 970 in wait_for_ready_callbacks
  File "C:\ci\ws\install\Lib\site-packages\rclpy\executors.py", line 1000 in _spin_once_impl
  File "C:\ci\ws\install\Lib\site-packages\rclpy\executors.py", line 1020 in spin_once
  File "C:\ci\ws\install\Lib\site-packages\launch_ros\ros_adapters.py", line 77 in _run
  File "C:\pixi_ws\.pixi\envs\default\Lib\threading.py", line 1010 in run
  File "C:\pixi_ws\.pixi\envs\default\Lib\threading.py", line 1073 in _bootstrap_inner
  File "C:\pixi_ws\.pixi\envs\default\Lib\threading.py", line 1030 in _bootstrap

Thread 0x0000247c (most recent call first):
  File "C:\pixi_ws\.pixi\envs\default\Lib\selectors.py", line 314 in _select
  File "C:\pixi_ws\.pixi\envs\default\Lib\selectors.py", line 323 in select
  File "C:\pixi_ws\.pixi\envs\default\Lib\asyncio\base_events.py", line 1949 in _run_once
  File "C:\pixi_ws\.pixi\envs\default\Lib\asyncio\base_events.py", line 641 in run_forever
  File "C:\ci\ws\install\Lib\site-packages\launch\utilities\signal_management.py", line 149 in run_background_loop
  File "C:\pixi_ws\.pixi\envs\default\Lib\threading.py", line 1010 in run
  File "C:\pixi_ws\.pixi\envs\default\Lib\threading.py", line 1073 in _bootstrap_inner
  File "C:\pixi_ws\.pixi\envs\default\Lib\threading.py", line 1030 in _bootstrap
....
Failed   <<< test_launch_ros [5min 30s, exited with code 3221225477]

This have been happening around 40% of the time, there are 7 failed builds last 15 days with this error

CC: @Crola1702

[knmcguire]

Isn't the operating system Windows 11 now that is on that runner

[mjcarroll]

I believe I have identified the root cause of these access violations. They appear to be caused by an unsafe shutdown pattern in several tests in test_launch_ros.

In these tests, ls.shutdown() was being skipped if the launch_task had already completed. However, the LaunchService may still have active resources, such as the background thread started by ROSAdapter. When ls.shutdown() is skipped, the LaunchService and its LaunchContext are destroyed, while the background thread continues to run and eventually attempts to access the now-destroyed context/node, leading to the 0xC0000005 access violation on Windows.

I have submitted a fix here: #542

[miguelgonrod]

Hi, this issue happened today again

• https://ci.ros2.org/job/nightly_win_rel/3815
• https://ci.ros2.org/job/nightly_win_rel/3814
• https://ci.ros2.org/job/nightly_win_rep/4119

[MiguelCompany]

eProsima/Fast-DDS#6503 has been merged; backports are on the way.

This means the issue is probably fixed for rolling and lyrical.

[miguelgonrod]

Hi today this failed in test_launch_ros: https://ci.ros2.org/job/nightly_win_rep/4202 and is running on rolling