How to use RSS right for many different scale sets in one cluster?

[granescb]

Hello everyone,

We are currently working on migrating from our old ARC system to RSS. Previously, we had various runners distinguished by their resources, sizes, and architectures (e.g., ARM). For instance, we had runners like small-arm64 and medium-amd64.

How can we replicate this setup using the new RSS approach? I attempted to use the original Helm chart for RSS installation, but it resulted in an excessive allocation of resources for just one RSS installation. If I need 10 different runners, I would end up with 10 Helm chart installations, each generating its Kubernetes resources (such as AutoscalingRunnerSet, Role, ServiceAccount, etc.).

To address this, we developed our own Helm chart for RSS installation. This allows us to create only one ServiceAccount and Role shared among all our AutoscalingRunnerSet resources and configure them through a single chart. However, we observed that each AutoscalingRunnerSet creates its own listener, which consumes 1 CPU and 2 GB of memory.

Previously, we had one controller, one webhook listener, and our runners. Now, we have one controller, 10 listeners, and the runners. How can we configure this correctly to achieve a similar setup with different runner types?