Add option to disable/remove autoscaling on node pools during turndown

[michaelmdresser]

If a cloud provider (GKE, EKS) autoscaler is present on node pools, turndown currently evicts pods and then relies on the autoscaler to reduce the node count. However, sometimes autoscaler parameters may be set with a too-high minimum node count, resulting in smaller than expected savings during turndown. This is a request to add an option to turndownschedules.kubecost.com to force turndown to behave like it does if an autoscaler isn't present.

This could be achieved by (1) modifying the minimum node count or (2) removing the autoscaling node pool entirely during turndown and recreating it on turnup.

[michaelmdresser]

See current behavior in turndown for reference, "Resize all the non-autoscaling node pools to 0":

cluster-turndown/pkg/turndown/turndown.go

Lines 224 to 245 in 28ab31c

	// 4. Filter out the current node pool holding the current node and/or autoscaling
	targetPools := []cp.NodePool{}
	for _, np := range nodePools {
	if np.Name() == currentNodePoolID || np.AutoScaling() {
	continue
	}
	targetPools = append(targetPools, np)
	}
	// Set NodePools on instance for resetting/upscaling
	ktdm.nodePools = targetPools
	ktdm.autoScaling = &isAutoScalingCluster
	ktdm.log.Log("Resizing all non-autoscaling node groups to 0...")
	// 5. Resize all the non-autoscaling node pools to 0
	err = ktdm.provider.SetNodePoolSizes(targetPools, 0)
	if err != nil {
	// TODO: Any steps that fail AFTER draining should revert the drain step?
	return err
	}

[Ayush9026]

Hi @michaelmdresser i want to work on this issue.

[michaelmdresser]

Hi @Ayush9026, you're welcome to investigate this behavior and open a pull request! I'll just warn you that we'll ask for thorough testing before we're willing to approve and merge. I think this issue's description is fairly thorough and provides at least a lead in to the code. Thanks for your interest.