Skip to main content

MarkLogic Server on Kubernetes

Huge pages

This section explains setting up and using huge pages. For additional information, see the Kubernetes documentation.

Warning

To increase performance and efficiency, disable transparent huge pages before following the steps in this section. Especially, on nodes with high memory utilization.

Set huge pages at the node level

Huge pages are configured by setting the kernel parameter vm.nr_hugepages. This parameter can be set using DaemonSet:

apiVersion: apps/v1
kind: DaemonSet
metadata:
  name: sysctl-hugepages
  namespace: kube-system
  labels:
    k8s-app: sysctl-hugepages
spec:
  selector:
    matchLabels:
      name: sysctl-hugepages
  template:
    metadata:
      labels:
        name: sysctl-hugepages
    spec:
      tolerations:
      # these tolerations are to have the daemonset runnable on control plane nodes
      # remove them if your control plane nodes should not run pods
      - key: node-role.kubernetes.io/control-plane
        operator: Exists
        effect: NoSchedule
      - key: node-role.kubernetes.io/master
        operator: Exists
        effect: NoSchedule
      containers:
      - name: sysctl
        image: busybox
        command: ["/bin/sh"]
        args: ["-c", "sysctl -w vm.hugetlb_shm_group=100; sysctl -w vm.nr_hugepages=1280; tail -f /dev/null"]
        securityContext:
          privileged: true
        resources:
          limits:
            memory: 200Mi
          requests:
            cpu: 100m
            memory: 200Mi
      terminationGracePeriodSeconds: 30
      # these nodeSelector is to have the daemonset only running on specific nodes hosting ML pods
      # adapt value if necessary
      nodeSelector:
        role: ml-worker

The docker image used is the standard busybox.

Arguments

Huge pages are set with these arguments:

  • args: ["-c", "sysctl -w vm.hugetlb_shm_group=100; sysctl -w vm.nr_hugepages=1280; tail -f /dev/null"]

  • vm.hugetlb_shm_group=100 (gid of default ml user)

  • vm.nr_hugepages=1280

Note

Linux huge pages should be set at 3/8 the size of physical memory.

Set privileged to true

Set the securityContext: privileged setting to true as shown below (and in Set huge pages at the node level).

containers:
      - name: sysctl
        image: busybox
        command: ["/bin/sh"]
        args: ["-c", "sysctl -w vm.hugetlb_shm_group=100; sysctl -w vm.nr_hugepages=1280; tail -f /dev/null"]
        securityContext:
          privileged: true
In this section: