Best practices for manual scaling

D-FKE supports manual scale-in and scale-out of worker nodes. For portal console instructions, see the user guide/.

Manual scale-in process detail

1. Pre-scale

• The system sets up the environment for cluster scaling.
• Checks environment: cluster status / network status.

2. Scale in

• Cordons all nodes (schedule disabled — prevents new application pods from entering the node).
• Drains each node in sequence (removes application pods from the current node and moves them to other nodes; nodes with local disks, single pods, disruption budgets, etc. cannot be removed — reference).
• After all nodes are drained, removes each node from the Kubernetes cluster.

3. Post-scale

• After all nodes are removed from the cluster, the system reclaims and deletes the VMs.
• Cleanup.

Manual scale-out process detail

1. Pre-scale

• The system sets up the environment for cluster scaling.
• Checks environment: cluster status / network status.
• The system creates VMs corresponding to the nodes to be added to the Kubernetes cluster.
• Sets up the environment to run Kubernetes on each VM.

2. Scale out

• Adds each node to the cluster in sequence.
• Allows application pod scheduling onto the newly joined nodes.

3. Post-scale

• Cleanup.

Recommendations when using manual scale

For scale-in operations:

• If scaling in many nodes, survey and size the remaining nodes to ensure they have sufficient resources for the application. Perform during low-load periods such as after business hours or on weekends.
• If you can identify the list of nodes to delete, cordon them in advance to prevent new workloads from entering nodes that will be deleted (based on monitoring node resource usage).
• If there is still a large workload on the node, pre-drain the node to be deleted to speed up the scale-in process and avoid timeouts.
• Split scale-in operations into smaller batches — perform no more than 10 nodes per manual scale-in.

For scale-out operations:

• Split scale-out operations into smaller batches — perform no more than 10 nodes per manual scale-out.

If the process times out:

• Press the retry action to re-run the process. See here/ for details.