Add support to parse multiple cores for setting CPU affinity

[saikishor]

This PR adds a new method that helps to set multiple CPUs for affinity instead of only one and this will help in choosing couple of CPUs to run the controller_manager etc

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 76.47059% with 4 lines in your changes missing coverage. Please review.

Project coverage is 72.72%. Comparing base (069df6d) to head (9ac21d4).

Files with missing lines	Patch %	Lines
src/realtime_helpers.cpp	76.47%	0 Missing and 4 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #208      +/-   ##
==========================================
+ Coverage   72.55%   72.72%   +0.17%     
==========================================
  Files           7        7              
  Lines         419      429      +10     
  Branches       68       71       +3     
==========================================
+ Hits          304      312       +8     
  Misses         73       73              
- Partials       42       44       +2     

Flag	Coverage Δ
unittests	72.72% <76.47%> (+0.17%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/realtime_helpers.cpp	39.79% <76.47%> (+4.56%)	⬆️

[firesurfer]

I just saw the PR by chance and was interested as I authored the original one. I am wondering in which (realtime) scenario would you like to prefer a set of cores? From my understanding every time a thread is executed on the different core we suffer from

a) Context switch overhead
b) Additional cache misses

Both brings in an uncertainty which we do not want in RT scenarios.

The only case I can imagine this could help is modern CPUs with multiple dies inside (e.g. most Ryzen processors). But the OS scheduler will already take care of trying not to move threads between cores on different dies per default.

[saikishor]

@firesurfer You can set it at the startup as we do in the ros2_control_node or in the configure of controllers and hardware components

[firesurfer]

@firesurfer You can set it at the startup as we do in the ros2_control_node or in the configure of controllers and hardware components

I am aware of that. But I just wanted to understand why it is an advantage in a RT scenario to use the group of cores instead of tying the thread to a single core.

[saikishor]

I am aware of that. But I just wanted to understand why it is an advantage in a RT scenario to use the group of cores instead of tying the thread to a single core.

By default, it uses mostly Core 0 as other. If you can use a set of performance cores, it would be better for the RT application, if your application needs more computational power than one core, then this definitely helps

[christophfroehlich]

LGTM

[christophfroehlich]

@Mergifyio backport humble

[mergify]

backport humble

✅ Backports have been created

• #223 Add support to parse multiple cores for setting CPU affinity (backport #208) has been created for branch humble