Re-evaluate Keda?

[hervenicol]

We have lots of sizing issues for Mimir and Loki.

I think if we could have both some automatic VPA and HPA based on different metrics (ie VPA based on RAM and CPU, HPA based on number of series), we could have a much more flexible autoscaling.

But this requires Keda. That we discarded, because nobody wants to assume/maintain it.

Given the time we spend manually sizing our components, maybe owning and maintaining Keda on the MCs would cost us less?