Introduction
This article provides answers to common questions about the process for scaling down a Crusoe Managed Kubernetes (CMK) nodepool.
Question 1: Why don't my nodepool VMs automatically delete when I lower the 'desired count'?
Answer
This is the intended, "safety-first" behavior of the platform. When you decrease the desired node count for a nodepool, CMK will honor this by preventing new nodes from being created, but it will not automatically delete existing VMs.
This choice is explicit. The system does not know which nodes might be running critical, single-replica workloads that you would not want evicted. To prevent accidental downtime, the platform gives you full control over which specific instances are terminated and when.
Question 2: What is the correct process to scale down a nodepool?
Answer
Scaling down a nodepool is a two-step process:
- Update the Nodepool's Desired Count: First, update your nodepool configuration using the Crusoe CLI. This sets a new "ceiling" for your nodepool.
-
Manually Terminate the Excess VMs: Next, you must go to the 'VM Instances' section of your project. From there, you can safely cordon and drain the workloads on the specific node(s) you want to remove using
kubectl, and then manually delete the VM instance(s) from the platform to match your new desired count.
Question 3: What happens if I lower the desired count but forget to delete the VMs?
Answer
The CMK control plane will respect the new, lower count and will not schedule new workloads on the extra nodes. However, the VM instances themselves will continue to run and incur billing charges. It is important to complete the manual deletion step to ensure your infrastructure and billing align with your desired state.
Additional Resources
For more details on managing nodepools, please see the official documentation: