Kubernetes autoscalers attempt to consolidate workloads to run the same applications on fewer nodes. However, when nodes contain unevictable Pods—Pods that cannot be safely evicted due to constraints such as Pod Disruption Budgets or eviction restrictions (for example, safe-to-evict settings)—those nodes cannot be drained.
As a result, Kubernetes clusters often suffer from Pod fragmentation, where Pods are spread inefficiently across nodes and unused capacity is distributed instead of consolidated.
These Pods become effectively pinned to nodes, blocking autoscaler scale-down.
Pod fragmentation directly limits the effectiveness of Pod Rightsizing and Headroom Reduction. Even when workloads are fully optimized, inefficient placement can prevent node consolidation, reducing the overall impact of these optimizations.
Adaptive Pod placement addresses this by improving how Pods are distributed across the cluster. It classifies Pods as either evictable or unevictable, and then adjusts their placement by grouping similar Pods and separating Pods of different classifications to reduce fragmentation and improve packing efficiency.
By enabling APP, you can:
Reduce fragmentation that blocks node scale-down.
Improve Pod packing across nodes.
Increase the effectiveness of the cluster autoscaler.
Enable more consistent and predictable workload placement.
To see your potential savings from using APP, run the Status check.
See also: