Decision for autoscaling with restricted downscaling

[simu]

No description provided.

[haasad]

Decision looks fine!

Two thoughts that came up while reviewing this:

• How big is the chance/risk of two nodes being drained simultaneously? One by the maintenance and one for downscaling. This could create problems for deployments without PDBs that rely on the fact the we usually never drain two nodes at the same time.
• Is there a risk that the autoscaler will always add a node during the maintenance (because resources are tight during node rebooting) which will then not be downscaled anymore (because the maintenance window is already over) and lay around until the next maintenance?

[simu]

How big is the chance/risk of two nodes being drained simultaneously? One by the maintenance and one for downscaling. This could create problems for deployments without PDBs that rely on the fact the we usually never drain two nodes at the same time.

There's a chance for this to happen, but I assume that if we enable downscaling at the start of the maintenance, any nodes that are getting scaled down will be gone by the time the maintenance updates the nodes.

Is there a risk that the autoscaler will always add a node during the maintenance (because resources are tight during node rebooting) which will then not be downscaled anymore (because the maintenance window is already over) and lay around until the next maintenance?

There is probably a risk for that, we could mitigate this by implementing a delay in the downscaling disable hook to allow the autoscaler to scale down such a node.