Using CPU Manager | Scaling and Performance Guide

What CPU Manager Does

CPU Manager is a Technology Preview feature.

CPU Manager manages groups of CPUs and constrains workloads to specific CPUs.

CPU Manager is useful for workloads that have some of these attributes:

Require as much CPU time as possible.
Are sensitive to processor cache misses.
Are low-latency network applications.
Coordinate with other processes and benefit from sharing a single processor cache.

Setting up CPU Manager

To set up CPU Manager:

Optionally, label a node:

# oc label node perf-node.example.com cpumanager=true

Enable CPU manager support on the target node:

# cat /etc/origin/node/node-config.yaml
...
kubeletArguments:
...
  feature-gates:
  - CPUManager=true
  cpu-manager-policy:
  - static
  cpu-manager-reconcile-period:
  - 5s
  kube-reserved: (1)
  - cpu=500m

# systemctl restart atomic-openshift-node

1	`kube-reserved` is a required setting. The value may need to be adjusted depending on your environment.

Create a pod that requests a core or multiple cores. Both limits and requests must have their CPU value set to a whole integer. That is the number of cores that will be dedicated to this pod:

# cat cpumanager.yaml
apiVersion: v1
kind: Pod
metadata:
  generateName: cpumanager-
spec:
  containers:
  - name: cpumanager
    image: gcr.io/google_containers/pause-amd64:3.0
    resources:
      requests:
        cpu: 1
        memory: "1G"
      limits:
        cpu: 1
        memory: "1G"
  nodeSelector:
    cpumanager: "true"

Create the pod:
```
# oc create -f cpumanager.yaml
```

Verify that the pod is scheduled to the node that you labeled:

# oc describe pod cpumanager
Name:         cpumanager-4gdtn
Namespace:    test
Node:         perf-node.example.com/172.31.62.105
...
    Limits:
      cpu:     1
      memory:  1G
    Requests:
      cpu:        1
      memory:     1G
...
QoS Class:       Guaranteed
Node-Selectors:  cpumanager=true
                 region=primary

Verify that the cgroups are set up correctly. Get the PID of the pause process:

# systemd-cgls -l
├─1 /usr/lib/systemd/systemd --system --deserialize 20
├─kubepods.slice
│ ├─kubepods-pod0ec1ab8b_e1c4_11e7_bb22_027b30990a24.slice
│ │ ├─docker-b24e29bc4021064057f941dc5f3538595c317d294f2c8e448b5e61a29c026d1c.scope
│ │ │ └─44216 /pause

Pods of QoS tier Guaranteed are placed within the kubepods.slice. Pods of other QoS tiers end up in child cgroups of kubepods.

# cd /sys/fs/cgroup/cpuset/kubepods.slice/kubepods-pod0ec1ab8b_e1c4_11e7_bb22_027b30990a24.slice/docker-b24e29bc4021064057f941dc5f3538595c317d294f2c8e448b5e61a29c026d1c.scope
# for i in `ls cpuset.cpus tasks` ; do echo -n "$i "; cat $i ; done
cpuset.cpus 2
tasks 44216

Check the allowed CPU list for the task:

# grep ^Cpus_allowed_list /proc/44216/status
Cpus_allowed_list:      2

Verify that another pod (in this case, the pod in the burstable QoS tier) on the system can not run on the core allocated for the Guaranteed pod:

# cat /sys/fs/cgroup/cpuset/kubepods.slice/kubepods-burstable.slice/kubepods-burstable-podbe76ff22_dead_11e7_b99e_027b30990a24.slice/docker-da621bea7569704fc39f84385a179923309ab9d832f6360cccbff102e73f9557.scope/cpuset.cpus
0-1,3

# oc describe node perf-node.example.com
...
Capacity:
 cpu:     4
 memory:  16266720Ki
 pods:    40
Allocatable:
 cpu:     3500m
 memory:  16164320Ki
 pods:    40
---
  Namespace                  Name                      CPU Requests  CPU Limits  Memory Requests  Memory Limits
  ---------                  ----                      ------------  ----------  ---------------  -------------
  test                        cpumanager-4gdtn          1 (28%)       1 (28%)     1G (6%)          1G (6%)
  test                        cpumanager-hczts          1 (28%)       1 (28%)     1G (6%)          1G (6%)
  test                        cpumanager-r9wrq          1 (28%)       1 (28%)     1G (6%)          1G (6%)
...
Allocated resources:
  (Total limits may be over 100 percent, i.e., overcommitted.)
  CPU Requests  CPU Limits  Memory Requests  Memory Limits
  ------------  ----------  ---------------  -------------
  3 (85%)       3 (85%)     5437500k (32%)   9250M (55%)

This VM has four CPU cores. You set kube-reserved to 500 millicores, meaning half of one core is subtracted from the total capacity of the node to arrive at the Node Allocatable amount.

You can see that Allocatable CPU is 3500 millicores. This means we can run three of our CPU manager pods since each will take one whole core. A whole core is equivalent to 1000 millicores.

If you try to schedule a fourth pod, the system will accept the pod, but it will never be scheduled:

# oc get pods --all-namespaces |grep test
test              cpumanager-4gdtn               1/1       Running            0          8m
test              cpumanager-hczts               1/1       Running            0          8m
test              cpumanager-nb9d5               0/1       Pending            0          8m
test              cpumanager-r9wrq               1/1       Running            0          8m