Does CAST AI have any general recommendations for EKS to speed up cluster scaling?

Let's explain this using an example scenario:

When a user scales 1000+ pods in parallel, they take around 5 to 8 minutes to get scheduled and running as the underlying node allocation is slower. Can CAST AI speed this up?

Here are a few solutions:

  1. Use Bottlerocket AMI.
  2. Move away from HPA to KEDA - it starts scaling sooner and removes big shocks later than when using HPA.
  3. Create headroom with low-priority fake pods (those could be spun on schedule if the timing is repeatable).
  4. Check if the team has some DaemonSets that take a long time to initialize the node to the Ready state.