Skip to content

Latest commit

 

History

History
94 lines (69 loc) · 3.9 KB

README.md

File metadata and controls

94 lines (69 loc) · 3.9 KB
Raw

Performance Tuning

All manifests should be applied automatically on a new cluster via make deploy, in case if you want to test each feature separately, refer to the feature section.

Before you apply any feature, you should:

  • add worker-rt role to relevant nodes oc label node <node_name> node-role.kubernetes.io/worker-rt=
  • add worker-rt machine config pool oc apply -f ../mcp/00-mcp-worker-rt.yaml

Environment variables

This is a list of environment variables that you should export before running make generate.

  • ISOLATED_CPUS - CPU's that you want to isolate from the system usage.
  • NON_ISOLATED_CPUS - CPU's that you want to reserve for OS system tasks.
  • RESERVED_CPUS - CPU's that you want to reserve for the system and does not use for containers workloads.
  • HUGEPAGES_NUMBER - Number of 1Gb hugepages to enable.
  • RT_REPO_URL - the location of a yum repo which provides the RT kernel RPMs as long as they are not part of RHCOS yet.
    Defaults to a RH internal URL. For deployments outside the RH network provide a yum repo on a reachable host and update this URL.
    To configure your own yum repo check the link.

Huge Pages

To verify huge pages feature functionality:

  • run make generate
  • apply huge pages kernel boot parameters config via oc create -f manifests/generated/10-machine-config-worker-rt-kargs.yaml
  • wait for workers update
oc wait machineconfigpools worker-rt --for condition=Updating --timeout=1800s
oc wait machineconfigpools worker-rt --for condition=Updated --timeout=1800s
  • create the huge pages pod via oc create -f examples/hugepages-pod.yaml
  • get pod UID by running oc get pods -o custom-columns=NAME:.metadata.name,UID:.metadata.uid,HOST_IP:.status.hostIP
  • now you can find the specific cgroup under the host where the pod is running and check hugepages limit cat /sys/fs/cgroup/hugetlb/kubepods.slice/kubepods-pod<pod_uid>.slice/hugetlb.1GB.limit_in_bytes
  • under the pod you can check the huge pages mount mount | grep hugepages

RT kernel

The realtime kernel will be installed using MachineConfig, which installs a new systemd unit, which runs a script. The template for the MachineConfig placed under manifests/templates and the script located in assets. The actual manifest is created by running make generate, which will base64 encode the script and put it into the template. The result is stored under manifests/generated/11-machine-config-worker-rt-kernel.yaml.

CPU manager

To enable the CPU manager, you should:

  • make generate
  • enable CPU static policy via KubeletConfiguration, oc create -f manifests/generated/12-kubeletconfig-worker-rt
  • wait for workers update
oc wait machineconfigpools worker-rt --for condition=Updating --timeout=1800s
oc wait machineconfigpools worker-rt --for condition=Updated --timeout=1800s

Topology Manager

To enable the topology manager, you should:

  • enable the topology manager feature gate oc replace -f manifests/12-feature-gate-latency-sensitive.yaml
  • wait for workers update
oc wait machineconfigpools worker-rt --for condition=Updating --timeout=1800s
oc wait machineconfigpools worker-rt --for condition=Updated --timeout=1800s
  • update the kubelet configuration oc apply -f manifests/12-kubeletconfig-worker-rt.yaml
  • wait for workers update
oc wait machineconfigpools worker-rt --for condition=Updating --timeout=1800s
oc wait machineconfigpools worker-rt --for condition=Updated --timeout=1800s
  • verify that kubelet configuration on the worker-rt updated with relevant parameters
cat /etc/kubernetes/kubelet.conf
{
    "kind":"KubeletConfiguration",
    ...
    "topologyManagerPolicy":"best-effort",
    "featureGates": {
        ...
        "TopologyManager":true,
        ...
    },
}