Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.nebius.com/llms.txt

Use this file to discover all available pages before exploring further.

In Managed Service for Kubernetes, the kubelet sets kubeReserved values on worker nodes to reserve resources for Kubernetes system components. This article describes the default values and how the memory reservation is calculated.

​
Default values

By default, worker nodes use the following kubeReserved values:
  • cpu: 100m
  • ephemeral-storage: 1Gi
  • memory: 512Mi
However, for worker nodes with GPUs, Managed Kubernetes calculates the memory reservation value from the size of RAM you select for the node. This helps account for a known Kubernetes issue where the kubelet may not observe MemoryPressure right away. For more information, see Node-pressure Eviction in the official Kubernetes documentation. The calculation uses the following tiered approach:
  • 255Mi for nodes with less than 1 GiB of RAM
  • 25% of the first 4 GiB
  • 20% of the next 4 GiB (up to 8 GiB)
  • 10% of the next 8 GiB (up to 16 GiB)
  • 6% of the next 112 GiB (up to 128 GiB)
  • 2% of RAM above 128 GiB
  • Plus 100Mi on every node to handle pod eviction

​
Examples

The following table shows the resulting memory value in kubeReserved for the selected RAM size in your GPU nodes:
Node RAMkubeReserved.memory
8 GiB1944Mi
16 GiB2763Mi
128 GiB9644Mi
256 GiB12266Mi

​
See also