In 4.13, BestEffort pod cpu affinity is set to reserved cpuset only. In 4.12 and earlier, it uses reserved cpuset + free isolated cpuset.

The following example uses AWS SNO m6i.4xlarge 16vCPU 64GB mem.

$ oc label node --all sno-pp=enabled
$ oc label mcp --all sno-pp=enabled
$ oc create -f - <<EOF
apiVersion: performance.openshift.io/v2
kind: PerformanceProfile
metadata:
  name: sno-pp
spec:
  cpu:
    reserved: 0-3
    isolated: 4-15
  machineConfigPoolSelector:
    sno-pp: enabled
  nodeSelector:
    node-role.kubernetes.io/master: ""
EOF
$ oc create -f - <<EOF
apiVersion: machineconfiguration.openshift.io/v1
kind: KubeletConfig
metadata:
  name: sno-pp
spec:
  machineConfigPoolSelector:
    matchLabels:
      sno-pp: enabled
  kubeletConfig:
    cpuManagerPolicy: static 
    cpuManagerReconcilePeriod: 5s
    reservedSystemCPUs: 0-3
EOF

$ oc new-project test
$ oc create deployment sleep-a --image=registry.access.redhat.com/ubi8/ubi -- /bin/bash -c "trap : TERM INT; sleep infinity & wait"
$ oc set resources deployment sleep-a --requests=cpu=1,memory=32Mi --limits=cpu=1,memory=32Mi
$ oc create deployment sleep-b --image=registry.access.redhat.com/ubi8/ubi -- /bin/bash -c "trap : TERM INT; sleep infinity & wait"

$ oc rsh deploy/sleep-a cat /proc/1/status | grep Cpus_allowed_list
4
$ oc rsh deploy/sleep-a cat /sys/fs/cgroup/cpuset/cpuset.cpus
4

$ oc rsh deploy/sleep-b cat /proc/1/status | grep Cpus_allowed_list
0-3
$ oc rsh deploy/sleep-b cat /sys/fs/cgroup/cpuset/cpuset.cpus
0-3,5-15

$ oc rsh deploy/sleep-b yes >/dev/null &

$ oc debug $NODE
# ps -C "yes" -L -o pid,lwp,pcpu,cpuid,time # cpuid shows 0, 1, 2, 3. No 5-15 used.

4.13.2

Always

1. Configure cluster with PerformaceProfile reverved/isolated cpuset, enable cpumanager
2. Deploy BestEffort pod
3. Look at process Cpus_allowed_list

Process cpu affinity is set to reserved cpuset

Process cpu affinity is set to reserved cpuset + free isolated cpuset, just same as cgroups cpuset.cpus