When I wrote the article, Aodh was not flexible enough to use aggregates in the way I present here. At Ussuri, this was still the case; I have not checked Victoria or Wallaby.
A little more information/ranting:
In a training course, I had been using Heat/Aodh/Gnocchi/Ceilometer to demonstrate autoscaling based on CPU utilization. My demo broke in Stein, and I decided to publish my efforts trying to make it work.
Eventually, I gave up on CPU utilization and used another metric (memory usage) to demonstrate autoscaling. Sadly, many of the old Heat autoscaling examples (https://opendev.org/openstack/heat-templates/src/branch/master/hot) are based on cpu_util and also don’t work anymore.
To add insult to injury, shortly after writing my article I found that the Gnocchi project was essentially abandoned. Luckily, other people seem to have picked up development again about a year later, in September 2020.