Added ScaledObject Support For Downscaler/Downtime-Replicas Annotation

[samuel-esp]

Motivation

Actually ScaledObject doesn't support "downscaler/downtime-replicas" annotation. See issue #96

Changes

• Refactored Keda ScaledObject resource class in order to support "downscaler/downtime-replicas" annotation
• Refactored scaler.py class in order to support "downscaler/downtime-replicas" annotation for Keda
• Made Keda ScaledObject more clear
• Added and Refactored Keda Unit Tests
• Refactored Docs

Tests done

• Unit Tests

TODO

• I've assigned myself to this PR

[jonathan-mayer]

Other than these two and me finding it a bit weird needing to specify the downtime replicas on the scaled objects deployments, ... this looks good to me

[samuel-esp]

Other than these two and me finding it a bit weird needing to specify the downtime replicas on the scaled objects deployments, ... this looks good to me

I see this concern and I strongly agree with you on a theoretical level, ScaledObject is basically a custom HPA and doesn't hold the concept of "replicas". However since the ScaledObject controls the Deployment/StatefulSet through the pause annotation, it will invalid the "downscaler/downtime-replicas" if set at Deployment/StatefulSet level. It must be clear to the user, as stated inside the documentation, that "downscaler/downtime-replicas" value should be the same inside both objects, otherwise it won't make sense

[jonathan-mayer]

Other than these two and me finding it a bit weird needing to specify the downtime replicas on the scaled objects deployments, ... this looks good to me

I see this concern and I strongly agree with you on a theoretical level, ScaledObject is basically a custom HPA and doesn't hold the concept of "replicas". However since the ScaledObject controls the Deployment/StatefulSet through the pause annotation, it will invalid the "downscaler/downtime-replicas" if set at Deployment/StatefulSet level. It must be clear to the user, as stated inside the documentation, that "downscaler/downtime-replicas" value should be the same inside both objects, otherwise it won't make sense

I am not too familiar with ScaledObjects but I think it would make more sense if we "excluded" the Workloads that are being managed by the ScaledObject and let the ScaledObject scale them according to the pause annotation we set on it when scaling down.

[samuel-esp]

Other than these two and me finding it a bit weird needing to specify the downtime replicas on the scaled objects deployments, ... this looks good to me

I see this concern and I strongly agree with you on a theoretical level, ScaledObject is basically a custom HPA and doesn't hold the concept of "replicas". However since the ScaledObject controls the Deployment/StatefulSet through the pause annotation, it will invalid the "downscaler/downtime-replicas" if set at Deployment/StatefulSet level. It must be clear to the user, as stated inside the documentation, that "downscaler/downtime-replicas" value should be the same inside both objects, otherwise it won't make sense

I am not too familiar with ScaledObjects but I think it would make more sense if we "excluded" the Workloads that are being managed by the ScaledObject and let the ScaledObject scale them according to the pause annotation we set on it when scaling down.

That's a good approach as well, I included this piece of advice inside the usage notes of the docs in the ScaledObject section

[jonathan-mayer]

If possible we should consider to automatically exclude the workloads managed by the ScaledObject.

[samuel-esp]

If possible we should consider to automatically exclude the workloads managed by the ScaledObject.

I'll try to address this inside one of the next releases