update in 2019 June: I managed to implement it, check here ;-) will write more on how it was done…
This note is my brain dump on getting ideas to do machine learning enabled optimized nova scheduler weighing.
By default, it simply weighted all existing weighers with weighing factor 1.0.
see ref first : https://www.slideshare.net/guptapeeyush1/presentation1-23249150
The weighing was called by:
By default it’s all weighers
get_weighed_objectsis doing this:
for i, weight in enumerate(weights):
obj = weighed_objs[i]
obj.weight += weigher.weight_multiplier() * weight
Below are mentioned subroutines…