OpenShift Container Platform

Tuning performance for hosted cluster nodes

To run high-performance workloads on hosted control planes on Red Hat OpenStack Platform (RHOSP), you can create a performance profile and deploy a tuned NodePool resource.

Prerequisites

You have RHOSP flavor that has the necessary resources to run your workload, including dedicated CPU, memory, and host aggregate information.
You have an RHOSP network that is attached to SR-IOV or DPDK-capable NICs. The network must be available to the project used by hosted clusters.

Procedure

Create a performance profile that meets your requirements in a file called perfprofile.yaml. For example:

Example performance profile in a config map

apiVersion: v1
kind: ConfigMap
metadata:
  name: perfprof-1
  namespace: clusters
data:
  tuning: |
    apiVersion: v1
    kind: ConfigMap
    metadata:
      name: cnf-performanceprofile
      namespace: "${HYPERSHIFT_NAMESPACE}"
    data:
      tuning: |
        apiVersion: performance.openshift.io/v2
        kind: PerformanceProfile
        metadata:
          name: cnf-performanceprofile
        spec:
          additionalKernelArgs:
            - nmi_watchdog=0
            - audit=0
            - mce=off
            - processor.max_cstate=1
            - idle=poll
            - intel_idle.max_cstate=0
            - amd_iommu=on
          cpu:
            isolated: "${CPU_ISOLATED}"
            reserved: "${CPU_RESERVED}"
          hugepages:
            defaultHugepagesSize: "1G"
            pages:
              - count: ${HUGEPAGES}
                node: 0
                size: 1G
          nodeSelector:
            node-role.kubernetes.io/worker: ''
          realTimeKernel:
            enabled: false
          globallyDisableIrqLoadBalancing: true

If you do not already have environment variables set for the HyperShift Operator namespace, isolated and reserved CPUs, and huge pages count, create them before applying the performance profile.

Apply the performance profile configuration by running the following command:
```
$ oc apply -f perfprof.yaml
```
If you do not already have a CLUSTER_NAME environment variable set for the name of your cluster, define it.
Set a node pool name environment variable by running the following command:
```
$ export NODEPOOL_NAME=$CLUSTER_NAME-cnf
```
Set a flavor environment variable by running the following command:
```
$ export FLAVOR="m1.xlarge.nfv"
```

Create a node pool that uses the performance profile by running the following command:

$ hcp create nodepool openstack \
  --cluster-name $CLUSTER_NAME \
  --name $NODEPOOL_NAME \
  --node-count 0 \
  --openstack-node-flavor $FLAVOR

Patch the node pool to reference the PerformanceProfile resource by running the following command:

$ oc patch nodepool -n ${HYPERSHIFT_NAMESPACE} ${CLUSTER_NAME} \
  -p '{"spec":{"tuningConfig":[{"name":"cnf-performanceprofile"}]}}' --type=merge

Scale the node pool by running the following command:

$ oc scale nodepool/$CLUSTER_NAME --namespace ${HYPERSHIFT_NAMESPACE} --replicas=1

Wait for the nodes to be ready:

Wait for the nodes to be ready by running the following command:

$ oc wait --for=condition=UpdatingConfig=True nodepool \
  -n ${HYPERSHIFT_NAMESPACE} ${CLUSTER_NAME} \
  --timeout=5m

Wait for the configuration update to finish by running the following command:

$ oc wait --for=condition=UpdatingConfig=False nodepool \
  -n ${HYPERSHIFT_NAMESPACE} ${CLUSTER_NAME} \
  --timeout=30m

Wait until all nodes are healthy by running the following command:
```
$ oc wait --for=condition=AllNodesHealthy nodepool \
  -n ${HYPERSHIFT_NAMESPACE} ${CLUSTER_NAME} \
  --timeout=5m
```
You can make an SSH connection into the nodes or use the oc debug command to verify performance configurations.