What if the autoscaler can be told via build configuration to spin up a different configuration of an AWS, DO, etc instance?
For AWS you could select a different image based on configuration in the .drone.yml.
For DO you could specifiy the size.
This is somewhat possible today. If your Drone installation supports both amd64 and arm64 architectures, for example, you could run two separate autoscalers. You could run one autoscaler to provision arm64 servers on aws, and a second to provision amd64 servers on gcp. The autoscaler has matching logic [1] to facilitate this.
There are plans to update the matching logic to look at the node [2] section of the yaml which is accessible to the autoscaler in the Labels field of the drone.Stage structure. You would need to run an autoscaler for each discrete node combination, which should be reasonable for most teams using this feature.
[1] https://github.com/drone/autoscaler/blob/master/engine/planner.go#L270
[2] https://docs.drone.io/user-guide/pipeline/nodes/
Thanks for the reply. I think that’s an acceptable workaround and glad to see there are plans for the future.