Help Center/ Cloud Container Engine/ FAQs/ Node Pool/ What Should I Do for Repeated Scale-Out During Node Provisioning?
Updated on 2026-06-27 GMT+08:00
View PDF
Share

What Should I Do for Repeated Scale-Out During Node Provisioning?

Symptom

When unscheduled pods have heterogeneous resource requirements, Autoscaler may over-provision nodes due to resource fragmentation. A second scale-out can be triggered while nodes from the first are still being provisioned.

Possible Cause

The Autoscaler's scale-out simulation does not match the actual scheduler behavior.
  • Initial calculation: During the first scale-out, Autoscaler uses a greedy algorithm to compute the theoretical minimum nodes required.
  • Node creation: After the first scale-out is triggered, the pending pods remain unschedulable during node creation.
  • Recalculation: Autoscaler reevaluates scale-out requirements, simulating pod placement onto pending nodes from the initial scale-out via the integrated scheduler framework.
  • Fragmentation: The scheduler's node-selection logic (based on resource utilization, among other factors) may fragment resources across the new nodes. Consequently, the initially provisioned nodes cannot accommodate the full batch of pods, forcing additional scale-out.

Solution

This behavior is expected based on the Autoscaler's design and does not require intervention.

Parent Topic: Node Pool