Looping during rollouts (Unexpected HPA Behavior)

[fernando-payface]

I'm having the following experience:

I have a single POD and a simple CPU-based HPA, whose current metric is always below the target.
Then I start a rollout, and Argo provisions a new canary POD, routes 10% of the traffic to this POD, and then pauses.
Shortly after, the HPA reports that the metric was above the target, and it provisions a new stable POD.
Then I have two stable PODs and one canary. The rollout remains paused, creating a loop of POD creation until I have several canary and stable PODs.
It only starts to return to normal when I promote everything and end the rollout, and I no longer have any canary PODs.
While there are canary PODs, even though they are always below the target, the HPA keeps indicating that it is above the target and provisions more PODs. According to my research, this is expected behavior, with suggestions of pausing the HPA during rollouts, or using fixed values ​​for the canary, etc.
But I can't accept that this is expected; I must be doing something wrong. Has anyone experienced this and can help me?
Thank you very much.

[kostis-codefresh]

I have several examples of Argo Rollouts and HPA here if you want to take a look https://github.com/kostis-codefresh/rollouts-autoscaling-example

You didn't specify if you use a traffic manager or not in your example.