ros2param test failures (cross-platform)

[jacobperron]

Bug report

Required Info:

• Operating System:
  • All
• Installation type:
  • source
• Version or commit hash:
  • c9e9e2f
• DDS implementation:
  • N/A

Steps to reproduce issue

See nightly failures:

• https://ci.ros2.org/view/nightly/job/nightly_linux_repeated/1836/testReport/junit/ros2param.src.ros2.ros2cli.ros2param.test.test_verb_dump/TestVerbDump/test_verb_dump/
• https://ci.ros2.org/view/nightly/job/nightly_linux_repeated/1836/testReport/junit/ros2param.src.ros2.ros2cli.ros2param.test.test_verb_dump/TestVerbDump/test_verb_dump_invalid_path/
• https://ci.ros2.org/view/nightly/job/nightly_linux_repeated/1836/testReport/junit/ros2param.src.ros2.ros2cli.ros2param.test.test_verb_dump/TestVerbDump/test_verb_dump_print/

Expected behavior

Tests pass

Actual behavior

Tests fail

Additional information

Roughly, it looks like the failures started on April 5th (based on the nightly Linux repeated build).

[jacobperron]

It looks like the failures are because not enough time is given for discover to happen between the test node and the CLI node (ie. the daemon).

If I switch the RMW implementation to rmw_connext_cpp, then I am able to reproduce reliably (I think because it takes longer for Connext to initialize).

[jacobperron]

#481 isn't the fix we were looking for. Now we can see a timeout happening during the newly added wait for discovery: https://ci.ros2.org/view/nightly/job/nightly_linux-aarch64_repeated/1158/testReport/ros2param.src.ros2.ros2cli.ros2param.test.test_verb_dump/TestVerbDump/test_verb_dump/

[jacobperron]

I can reproduce the timeouts locally by first starting the daemon with an RMW implementation that is different than the one used by the tests:

RMW_IMPLEMENTATION=rmw_connext_cpp ros2 daemon start

And then running the tests

colcon test --packages-select ros2param

[hidmic]

Aha! That's why most ros2cli launch tests systematically stop that daemon e.g. https://github.com/ros2/ros2cli/blob/master/ros2topic/test/test_cli.py#L116.

And #389 never landed. I can go back and update it to replace these if you'd like.

[jacobperron]

@hidmic I noticed that #389 doesn't touch test_verb_dump.py, so I started converting that set of tests to launch tests, independent of #389.