SLCM latency tests fail when using the PerformanceProfile recommended for tier-1 partner in the Blueprint document. The profile is missing a required kubelet configuration for enabling certain sysctls, which are necessary for the latency tests to pass.    

OpenShift version: 4.16

Always

Deploy a Single Node OpenShift (SNO) using the following PerformanceProfile:
yamlCopyEditkind: PerformanceProfile
apiVersion: "performance.openshift.io/v2"
metadata:
  name: sno-perf-profile
spec:
  cpu:
    isolated: "2-55,58-111"
    reserved: "0,1,56,57"
  hugepages:
    pages:
      - size: "1G"
        count: 52
        node: 0
      - size: "1G"
        count: 52
        node: 1
  numa:
    topologyPolicy: restricted
  realTimeKernel:
    enabled: true
  workloadHints:
    highPowerConsumption: false
    perPodPowerManagement: false
    realTime: true
  nodeSelector:
    node-role.kubernetes.io/master: ""
  machineConfigPoolSelector:
    pools.operator.machineconfiguration.openshift.io/master: ""



Run SLCM latency test suite as defined.


Observe that the latency tests fail due to missing sysctl configuration.

Latency tests fail with the default profile provided in the Blueprint.

    All latency tests should pass when using the Blueprint-recommended PerformanceProfile.

# Fix: Add the following annotation to the PerformanceProfile to allow TCP socket reuse for pods on nodes with isolated CPUs:   
annotations:
  kubeletconfig.experimental: |
    {"allowedUnsafeSysctls":["net.ipv4.tcp_tw_reuse"]}