Manage HugePages

Configure and manage huge pages as a schedulable resource in a cluster.
FEATURE STATE: Kubernetes v1.28 [stable]

Kubernetes supports the allocation and consumption of pre-allocated huge pages by applications in a Pod. This page describes how users can consume huge pages.

Before you begin

Kubernetes nodes must pre-allocate huge pages in order for the node to report its huge page capacity.

A node can pre-allocate huge pages for multiple sizes, for instance, the following line in /etc/default/grub allocates 2*1GiB of 1 GiB and 512*2 MiB of 2 MiB pages:

GRUB_CMDLINE_LINUX="hugepagesz=1G hugepages=2 hugepagesz=2M hugepages=512"

The nodes will automatically discover and report all huge page resources as schedulable resources.

When you describe the Node, you should see something similar to the following in the following in the Capacity and Allocatable sections:

Capacity:
  cpu:                ...
  ephemeral-storage:  ...
  hugepages-1Gi:      2Gi
  hugepages-2Mi:      1Gi
  memory:             ...
  pods:               ...
Allocatable:
  cpu:                ...
  ephemeral-storage:  ...
  hugepages-1Gi:      2Gi
  hugepages-2Mi:      1Gi
  memory:             ...
  pods:               ...

API

Huge pages can be consumed via container level resource requirements using the resource name hugepages-<size>, where <size> is the most compact binary notation using integer values supported on a particular node. For example, if a node supports 2048KiB and 1048576KiB page sizes, it will expose a schedulable resources hugepages-2Mi and hugepages-1Gi. Unlike CPU or memory, huge pages do not support overcommit. Note that when requesting hugepage resources, either memory or CPU resources must be requested as well.

A pod may consume multiple huge page sizes in a single pod spec. In this case it must use medium: HugePages-<hugepagesize> notation for all volume mounts.

apiVersion: v1
kind: Pod
metadata:
  name: huge-pages-example
spec:
  containers:
  - name: example
    image: fedora:latest
    command:
    - sleep
    - inf
    volumeMounts:
    - mountPath: /hugepages-2Mi
      name: hugepage-2mi
    - mountPath: /hugepages-1Gi
      name: hugepage-1gi
    resources:
      limits:
        hugepages-2Mi: 100Mi
        hugepages-1Gi: 2Gi
        memory: 100Mi
      requests:
        memory: 100Mi
  volumes:
  - name: hugepage-2mi
    emptyDir:
      medium: HugePages-2Mi
  - name: hugepage-1gi
    emptyDir:
      medium: HugePages-1Gi

A pod may use medium: HugePages only if it requests huge pages of one size.

apiVersion: v1
kind: Pod
metadata:
  name: huge-pages-example
spec:
  containers:
  - name: example
    image: fedora:latest
    command:
    - sleep
    - inf
    volumeMounts:
    - mountPath: /hugepages
      name: hugepage
    resources:
      limits:
        hugepages-2Mi: 100Mi
        memory: 100Mi
      requests:
        memory: 100Mi
  volumes:
  - name: hugepage
    emptyDir:
      medium: HugePages
  • Huge page requests must equal the limits. This is the default if limits are specified, but requests are not.
  • Huge pages are isolated at a container scope, so each container has own limit on their cgroup sandbox as requested in a container spec.
  • EmptyDir volumes backed by huge pages may not consume more huge page memory than the pod request.
  • Applications that consume huge pages via shmget() with SHM_HUGETLB must run with a supplemental group that matches proc/sys/vm/hugetlb_shm_group.
  • Huge page usage in a namespace is controllable via ResourceQuota similar to other compute resources like cpu or memory using the hugepages-<size> token.
Last modified October 18, 2023 at 1:15 PM PST: Update scheduling-hugepages.md (ca2b3023e7)