Description

This closes ros2/rmw_zenoh#852. rmw_zenoh failed to pass the test since the publisher declared in test_async_sub was accidentally detected by the subsequent test_pub_sub (the tests run in alphabetical order). The root cause is that the publisher node is not explicitly destroyed during the rclpy shutdown.

To clarify, this behavior stems from the different mechanism used in rmw_zenoh compared to other DDS implementations. Specifically, the shutdown of rmw_zenoh doesn't immediately close the Zenoh session as long as other nodes are still alive. This aligns with the RMW philosophy that nodes and their associated entities such as publishers, subscriptions, and so on, are not explicitly destroyed by rmw_shutdown.

Introduing a sleep after rclpy.shutdown does not solve the issue because the node data remain referenced until the object destruction phase. Meanwhile, the next test begins immediately and reads the stale liveliness token, causing the test to fail.

This fix adotps the approach used in other tests by explicitly cleaning up the nodes to avoid such stale references.

Is this user-facing behavior change?

Did you use Generative AI?

Additional Information

• Tested on rmw_zenoh 15915cb28fd732e76b269eab90e87c8f0c13f0a5

  ---

  This is an automatic backport of pull request fix(test_events_executor): destroy all nodes before shutdown #1538 done by Mergify.