👩‍🌾 test_info_by_topic flaky on Linux

[chapulina]

The test is flaky for CycloneDDS on Linux aarch64 (failed 6/24 latest runs):

https://ci.ros2.org/view/nightly/job/nightly_linux-aarch64_repeated/1398/testReport/projectroot/test/test_info_by_topic__rmw_cyclonedds_cpp/history/

It's also slightly flaky for FastRTPS on aarch64 (failed 1/24 latest runs):

https://ci.ros2.org/view/nightly/job/nightly_linux-aarch64_repeated/1398/testReport/projectroot/test/test_info_by_topic__rmw_fastrtps_cpp/history/

On x86_64, it passes consistently for CycloneDDS:

https://ci.ros2.org/view/nightly/job/nightly_linux_repeated/2096/testReport/projectroot/test/test_info_by_topic__rmw_cyclonedds_cpp/history/

But it's slightly flaky for FastRTPS (failed 1/24 latest runs):

https://ci.ros2.org/view/nightly/job/nightly_linux_repeated/2096/testReport/projectroot/test/test_info_by_topic__rmw_fastrtps_cpp/history/

Failure log for the flakiest case:

[ RUN      ] TestInfoByTopicFixture__rmw_cyclonedds_cpp.test_rcl_get_publishers_subscription_info_by_topic
/home/jenkins-agent/workspace/nightly_linux-aarch64_repeated/ws/src/ros2/rcl/rcl/test/rcl/test_info_by_topic.cpp:367: Failure
Expected equality of these values:
  topic_endpoint_info_array_pub.size
    Which is: 0
  1u
    Which is: 1
Expected one publisher
[  FAILED  ] TestInfoByTopicFixture__rmw_cyclonedds_cpp.test_rcl_get_publishers_subscription_info_by_topic (16 ms)

[fujitatomoya]

adding extra sleep in here would be okay enough, or maybe we can call wait_for_established_subscription instead.

[fujitatomoya]

#859 should fix the flaky situation. I tried more than 30 times but i was unable to reproduce this issue.

[fujitatomoya]

@chapulina
CC: @clalancette

friendly ping.

[fujitatomoya]

@chapulina
CC: @clalancette

i think we can close this one, since #859 is merged.

[chapulina]

Thanks. Closing for now, please reopen if the test starts flaking again.

[Blast545]

This is happening in the Foxy CI jobs: https://build.ros2.org/view/Fci/job/Fci__nightly-cyclonedds_ubuntu_focal_amd64/. 18/100 of the last jobs.

Didn't check the other Foxy jobs, but it looks like the fix addressing this issue wasn't backported to Foxy. Should we do something about it? @jacobperron @clalancette

[clalancette]

Didn't check the other Foxy jobs, but it looks like the fix addressing this issue wasn't backported to Foxy. Should we do something about it? @jacobperron @clalancette

I don't see any problem doing that; the fix is strictly to the tests, so there are no API/ABI concerns.

[jacobperron]

I've opened a backport PR: #944