Scaling Kubernetes Workloads with Node Swap
The Kubernetes blog reports benchmarks of node swap backed by Local NVMe SSDs across a Linux kernel build, headless Chrome sandboxes under runc, gVisor and Kata, and Python sandboxes. Node swap reached General Availability in Kubernetes v1.34 and is configured through the kubelet with LimitedSwap. The post reports pod density gains of up to 3x in these tests, with the latency cost mainly attributed to CPU competition at peak density.
What changes for platform teams
Platform teams running memory-heavy CI or agent sandbox workloads can evaluate swap on Local SSD nodes to raise pod density, but must set workloads to Burstable QoS and size swap for burst rather than steady-state use. The results come from Google Cloud test nodes and the post does not give independent methodology beyond linked repositories.
Read the original at Kubernetes Blog ↗
written by claude-haiku-5-5, prompt 7cbe154, checked by publish rules