[backport][ros2] make transient local reliable (#1033)

[Karsten1987]

eloquent backport of #1033

[Karsten1987]

test failures seem unrelated to the fix and the newly introduced test ran successfully.

[chapulina]

Actually it looks like the new test is responsible for the test failures. It looks like gzserver didn't shut down properly at the end of the test:

11:32:38 19: [INFO] [gzserver-2]: sending signal 'SIGINT' to process[gzserver-2]
11:32:38 19: [INFO] [mock_robot_state_publisher-1]: sending signal 'SIGINT' to process[mock_robot_state_publisher-1]
11:32:38 19: [mock_robot_state_publisher-1] [INFO] [mock_robot_state_publisher]: Publishing on /mock_robot_description
11:32:38 19: [mock_robot_state_publisher-1] [INFO] [rclcpp]: signal_handler(signal_value=2)
11:32:38 19: [INFO] [mock_robot_state_publisher-1]: process has finished cleanly [pid 7382]
11:32:43 19: [ERROR] [gzserver-2]: process[gzserver-2] failed to terminate '5' seconds after receiving 'SIGINT', escalating to 'SIGTERM'
11:32:43 19: [INFO] [gzserver-2]: sending signal 'SIGTERM' to process[gzserver-2]
11:32:43 19: [ERROR] [gzserver-2]: process has died [pid 7383, exit code -15, cmd 'gzserver                                                                    -s libgazebo_ros_init.so   -s libgazebo_ros_factory.so       '].

Which makes all subsequent tests fail with

11:32:46 7: �[1;31m[Err] [Master.cc:96] �[0m�[1;31mEXCEPTION: �[0m�[1;31mUnable to start server[bind: Address already in use]. There is probably another Gazebo process running.
11:32:46 7: �[0m�[1;31m
11:32:46 7: �[0m�[1;31m�[0mterminate called after throwing an instance of 'gazebo::common::Exception'

Could be flaky, let's try again.

[chapulina]

@ros-pull-request-builder retest this please

[chapulina]

Same thing again. I wonder if it's somehow related to the CI infrastructure, since #1033 tested ok on ci.ros2.org.

I'm adding PR CI for eloquent to the OSRF farm so we have another point of reference: https://bitbucket.org/osrf/release-tools/pull-requests/945/

[chapulina]

@osrf-jenkins run tests please

[hidmic]

It looks like gzserver didn't shut down properly at the end of the test:

Maybe 5 seconds isn't enough for Gazebo to shut down? It'd be a bit unfortunate, but you can try bumping up SIGTERM and SIGKILL timeouts.

[chapulina]

Thanks, @hidmic. A combination of the timeouts and #1037 worked locally for me: 36e9bd3

[chapulina]

Thanks for the endurance