roscpp ActionClient subscription queue size should be consistent with rospy.

[fujitatomoya]

related to #144

roscpp ActionClient subscription queue size should be more than 1, and also consistent with rospy.

Signed-off-by: Tomoya.Fujita Tomoya.Fujita@sony.com

[fujitatomoya]

i am not sure what went wrong with my fix, all i see RESULT: SUCCESS from the CI log. could anyone help me out?

[StefanFabian]

Most likely not your fault.
One of the tests runs out of time:
http://build.ros.org/job/Npr__actionlib__ubuntu_focal_amd64/23/testReport/rostest.runner/RosTest/testtest_cpp_simple_client/

Here's the test:
https://github.com/ros/actionlib/blob/noetic-devel/actionlib/test/exercise_simple_client.cpp

One possible solution would be to increase the time limit here:
https://github.com/ros/actionlib/blob/noetic-devel/actionlib/test/test_exercise_simple_clients.launch

[fujitatomoya]

how do we kick start the CI once again?
/CI

[StefanFabian]

I assume only maintainers can trigger a rebuild.
Maybe @mjcarroll can help?

[mjcarroll]

@ros-pull-request-builder retest this please

[mjcarroll]

It looks like we're still getting a timeout here.

@fujitatomoya would you like to try to expand the timeout as @StefanFabian suggested?

[fujitatomoya]

according to testlog, 90 seconds gives it enough buffer.

[fujitatomoya]

@mjcarroll @StefanFabian

i dont think CI test failures are related to my fix.
http://build.ros.org/job/Npr__actionlib__ubuntu_focal_amd64/25/testReport/junit/(root)/SimpleClientFixture/just_succeed/
http://build.ros.org/job/Npr_db__actionlib__debian_buster_amd64/25/testReport/junit/(root)/SimpleClientFixture/just_succeed/

@ros-pull-request-builder retest this please

[fujitatomoya]

@ros-pull-request-builder retest this please

[StefanFabian]

according to testlog, 90 seconds gives it enough buffer.

I don't know where you get that from since the test is preempted after 60 seconds without information about how much longer it might take.
However, the new error does indicate that time may not be the problem.
The problem is that the client is still in the ACTIVE state after 60 seconds.
The question is whether that is because of errors in the communication, i.e., the server is finished but the client doesn't know about it, or if the server is actually not finished yet.
Have you tried running the tests locally?

[fujitatomoya]

@StefanFabian

Have you tried running the tests locally?

of course, i have.

root@9b04ce524e20:~/ros_ws/catkin_ws# catkin_make run_tests actionlib
...<snip>
[actionlib.rosunit-test_cpp_simple_client/just_succeed][passed]
[actionlib.rosunit-test_cpp_simple_client/just_abort][passed]
[actionlib.rosunit-test_cpp_simple_client/just_preempt][passed]
[actionlib.rosunit-test_cpp_simple_client/drop][passed]
[actionlib.rosunit-test_cpp_simple_client/exception][passed]
[actionlib.rosunit-test_cpp_simple_client/ignore_cancel_and_succeed][passed]
[actionlib.rosunit-test_cpp_simple_client/lose][passed]
[actionlib.rosunit-test_py_simple_client/test_drop][passed]
[actionlib.rosunit-test_py_simple_client/test_exception][passed]
[actionlib.rosunit-test_py_simple_client/test_ignore_cancel_and_succeed][passed]
[actionlib.rosunit-test_py_simple_client/test_just_abort][passed]
[actionlib.rosunit-test_py_simple_client/test_just_preempt][passed]
[actionlib.rosunit-test_py_simple_client/test_just_succeed][passed]
[actionlib.rosunit-test_py_simple_client/test_lose][passed]

SUMMARY
 * RESULT: SUCCESS
 * TESTS: 14
 * ERRORS: 0
 * FAILURES: 0
...<snip>

[fujitatomoya]

according to the test history,
http://build.ros.org/job/Npr__actionlib__ubuntu_focal_amd64/27/testReport/junit/(root)/SimpleClientFixture/just_succeed/history/
http://build.ros.org/job/Npr_db__actionlib__debian_buster_amd64/27/testReport/(root)/SimpleClientFixture/just_succeed/history/

client gets timeout via waitForResult, and goal is still in ACTIVE state because server does not finish the goal and never sends back the result. I am not sure why server is not responding but connected to the client since this is not reproducible in my environment.

I am negative on expanding timeout for the test, so reverting that commit.

[fujitatomoya]

let's see what's gonna happen at this time,

@ros-pull-request-builder retest this please

[StefanFabian]

Perhaps both the timeout of the test and the timeout of the waitForResult need to be increased?
I've had a look at the tests and in theory, the tests could be completed in ~20 seconds if there was no execution time except for waiting.
I would assume that even a python action server should not take more than a second to handle a request which would result in at most 30 seconds of runtime even on limited hardware but who knows.
Maybe you should try increasing the time limit of the test to 3 minutes and the client wait timeout to 120 seconds.
I would keep the test timeout greater than the client wait timeout because the error message that the server didn't answer in time is a lot more helpful than the message that the test was preempted.

[fujitatomoya]

@StefanFabian

alright, let's give it a shot.

[fujitatomoya]

as expected, does not fix the test failure. Server side still in active state, i certainly need to access the CI system to debug this problem to check what is happening on server side. any thoughts? BTW, i will revert the be7c705 cz this was just a trial.

[fujitatomoya]

@ros-pull-request-builder retest this please

[fujitatomoya]

still unstable, I do not know what to do. any thoughts?

[StefanFabian]

Without being able to debug, I can only make wild guesses.
You could modify the tests to add more logging and give each of the tests a unique result code so each call can be traced in the logs..
I'm seeing a lot of TCP/IP connection failures, not sure if that's relevant.

[fujitatomoya]

@ros-pull-request-builder retest this please

[mjcarroll]

LGTM, thanks for the contribution and trying to figure out the flaky test issues.