Kubernetes Blog

The article explains how to scale Kubernetes workloads using nodes with swap enabled to mitigate memory pressure from agentic AI workloads.

• •Memory often runs out before CPU, and agentic AI workloads demand large memory footprints to start up.
• •Idle but resident memory is expensive and limits the number of pods a node can hold.
• •Kubernetes support for running nodes with swap enabled has reached General Availability.

As Kubernetes adoption grows, the focus has shifted to managing complex application lifecycles, highlighting the importance of SIG Apps.

• •Modern platforms support stateless web services, stateful databases, batch processing, and AI workloads.
• •They must remain reliable during upgrades, scaling events, and infrastructure failures.
• •Every Kubernetes user relies on SIG Apps, whether they realize it or not.

In Kubernetes v1.37, key Workload-Aware Scheduling (WAS) features were promoted to Beta, strengthening the foundation for AI/ML workloads.

• •The Workload and PodGroup APIs, which enable gang scheduling, were promoted to Beta.
• •Workload-Aware Preemption (WAP) is now available as a Beta feature.
• •The shared DRA ResourceClaims feature for PodGroups also reached Beta.