Use spin() in component_manager_isolated.hpp

[alsora]

Replace spin_until_future_complete with spin in component_manager_isolated.hpp.

spin_until_future_complete() is more inefficient as it needs to check the state of the future and the timeout after every work iteration.

This PR addresses the race-condition issues mentioned in #1781 by using the recently added is_spinning() API

Tests on the navigation stack show 10% less CPU usage when using spin() rather than spin_until_future_complete()

FYI @SteveMacenski @gezp @ivanpauno

[fujitatomoya]

@alsora

Tests on the navigation stack show 10% less CPU usage when using spin() rather than spin_until_future_complete()

sounds great, would you mind sharing the platform you benchmarked this?

[alsora]

@fujitatomoya the tests have been carried out by @gezp, he can definitely provide you more details but I think the 10% less cpu was on a raspberry pi 4, while on x86 the improvement was ~5%

[fujitatomoya]

Rasp4 is one of the common benchmark platform we have too, this is good information. thanks!

[ivanpauno]

LGTM with green CI

[alsora]

CI

• Linux
• Linux-aarch64
• Windows

[alsora]

Hi, can someone give a look at the windows CI above? It says that there are more than 70 test failures.

They all seem unrelated tests to what I'm doing here, but they are really too many to ignore.

Looking at the logs it may also seem an infrastructure issue, with some tests unable to start.
Do you have any insight?

[ivanpauno]

Same failures are happening on nightlies, e.g.: https://ci.ros2.org/view/nightly/job/nightly_win_rel/2207/testReport/junit/(root)/projectroot/test_action_client/.

[ivanpauno]

Ignoring the windows pytest issues that also happen on nightlies, going in!