Extra Requirements in GMM Trainings

[Atticus1806]

After looking at some crashed jobs I found some lets say unusual values for requirements like GB in my GMM jobs (10, 7, 5).
Short investigation found that the parameters "*_extra_rqmt" do not replace the requirement, but rather add the requirements to the basic rqmt (c.f. udpate_rqmt in mm_sequence.py).
After talking to @JackTemaki we conlcuded that this is somewhat surprising and usually done differently, where handin rqmt to a job in the construction would replace the rqmt not add it.

In general this is no big deal, but for me the values in the baseline rather look like intended replacement values and not added values I just wanted to bring this up, because we might be able to save resources in the future with reducing some of them.

[michelwi]

I think that is why it is called *_extra_rqmt and not *_rqmt.

I was fully aware of that peculiarity, but agree that this could be surprising. Maybe we add the same parameters but without "extra" and those set the given value?

[Atticus1806]

I agree that the naming is somewhat clear. Maybe adding a docstring would help.
Still I wanted to bring to attention that the default added values like for LS100 is 8, which to me rather seems like this was inteded to be 8 in total.

For example for the ls960 pipeline this value is 6, so here it might be "correct", or is this difference intended?

Maybe @christophmluscher or @JackTemaki can comment?

In general this issue was not about a faulty behavior or necessarily a change, but just to start a discussion on whether this behavior is known/intended.