Control CPU Management Policies on the Node

Kubernetes keeps many aspects of how pods execute on nodes abstracted from the user. This is by design.  However, some workloads require stronger guarantees in terms of latency and/or performance in order to operate acceptably. The kubelet provides methods to enable more complex workload placement policies while keeping the abstraction free from explicit placement directives.

For detailed information on resource management, please refer to the Resource Management for Pods and Containers documentation.

Before you begin

You need to have a Kubernetes cluster, and the kubectl command-line tool must be configured to communicate with your cluster. It is recommended to run this tutorial on a cluster with at least two nodes that are not acting as control plane hosts. If you do not already have a cluster, you can create one by using minikube or you can use one of these Kubernetes playgrounds:

• Killercoda
• Play with Kubernetes

If you are running an older version of Kubernetes, please look at the documentation for the version you are actually running.

CPU Management Policies

By default, the kubelet uses CFS quota to enforce pod CPU limits.  When the node runs many CPU-bound pods, the workload can move to different CPU cores depending on whether the pod is throttled and which CPU cores are available at scheduling time. Many workloads are not sensitive to this migration and thus work fine without any intervention.

However, in workloads where CPU cache affinity and scheduling latency significantly affect workload performance, the kubelet allows alternative CPU management policies to determine some placement preferences on the node.

Configuration

The CPU Manager policy is set with the --cpu-manager-policy kubelet flag or the cpuManagerPolicy field in KubeletConfiguration. There are two supported policies:

• none: the default policy.
• static: allows pods with certain resource characteristics to be granted increased CPU affinity and exclusivity on the node.

The CPU manager periodically writes resource updates through the CRI in order to reconcile in-memory CPU assignments with cgroupfs. The reconcile frequency is set through a new Kubelet configuration value --cpu-manager-reconcile-period. If not specified, it defaults to the same duration as --node-status-update-frequency.

The behavior of the static policy can be fine-tuned using the --cpu-manager-policy-options flag. The flag takes a comma-separated list of key=value policy options. If you disable the CPUManagerPolicyOptions feature gate then you cannot fine-tune CPU manager policies. In that case, the CPU manager operates only using its default settings.

In addition to the top-level CPUManagerPolicyOptions feature gate, the policy options are split into two groups: alpha quality (hidden by default) and beta quality (visible by default). The groups are guarded respectively by the CPUManagerPolicyAlphaOptions and CPUManagerPolicyBetaOptions feature gates. Diverging from the Kubernetes standard, these feature gates guard groups of options, because it would have been too cumbersome to add a feature gate for each individual option.

Changing the CPU Manager Policy

Since the CPU manager policy can only be applied when kubelet spawns new pods, simply changing from "none" to "static" won't apply to existing pods. So in order to properly change the CPU manager policy on a node, perform the following steps:

1. Drain the node.
2. Stop kubelet.
3. Remove the old CPU manager state file. The path to this file is /var/lib/kubelet/cpu_manager_state by default. This clears the state maintained by the CPUManager so that the cpu-sets set up by the new policy won’t conflict with it.
4. Edit the kubelet configuration to change the CPU manager policy to the desired value.
5. Start kubelet.

Repeat this process for every node that needs its CPU manager policy changed. Skipping this process will result in kubelet crashlooping with the following error:

could not restore state from checkpoint: configured policy "static" differs from state checkpoint policy "none", please drain this node and delete the CPU manager checkpoint file "/var/lib/kubelet/cpu_manager_state" before restarting Kubelet

None policy

The none policy explicitly enables the existing default CPU affinity scheme, providing no affinity beyond what the OS scheduler does automatically.  Limits on CPU usage for Guaranteed pods and Burstable pods are enforced using CFS quota.

Static policy

The static policy allows containers in Guaranteed pods with integer CPU requests access to exclusive CPUs on the node. This exclusivity is enforced using the cpuset cgroup controller.

This policy manages a shared pool of CPUs that initially contains all CPUs in the node. The amount of exclusively allocatable CPUs is equal to the total number of CPUs in the node minus any CPU reservations by the kubelet --kube-reserved or --system-reserved options. From 1.17, the CPU reservation list can be specified explicitly by kubelet --reserved-cpus option. The explicit CPU list specified by --reserved-cpus takes precedence over the CPU reservation specified by --kube-reserved and --system-reserved. CPUs reserved by these options are taken, in integer quantity, from the initial shared pool in ascending order by physical core ID.  This shared pool is the set of CPUs on which any containers in BestEffort and Burstable pods run. Containers in Guaranteed pods with fractional CPU requests also run on CPUs in the shared pool. Only containers that are both part of a Guaranteed pod and have integer CPU requests are assigned exclusive CPUs.

As Guaranteed pods whose containers fit the requirements for being statically assigned are scheduled to the node, CPUs are removed from the shared pool and placed in the cpuset for the container. CFS quota is not used to bound the CPU usage of these containers as their usage is bound by the scheduling domain itself. In others words, the number of CPUs in the container cpuset is equal to the integer CPU limit specified in the pod spec. This static assignment increases CPU affinity and decreases context switches due to throttling for the CPU-bound workload.

Consider the containers in the following pod specs:

spec:
  containers:
  - name: nginx
    image: nginx

The pod above runs in the BestEffort QoS class because no resource requests or limits are specified. It runs in the shared pool.

spec:
  containers:
  - name: nginx
    image: nginx
    resources:
      limits:
        memory: "200Mi"
      requests:
        memory: "100Mi"

The pod above runs in the Burstable QoS class because resource requests do not equal limits and the cpu quantity is not specified. It runs in the shared pool.

spec:
  containers:
  - name: nginx
    image: nginx
    resources:
      limits:
        memory: "200Mi"
        cpu: "2"
      requests:
        memory: "100Mi"
        cpu: "1"

The pod above runs in the Burstable QoS class because resource requests do not equal limits. It runs in the shared pool.

spec:
  containers:
  - name: nginx
    image: nginx
    resources:
      limits:
        memory: "200Mi"
        cpu: "2"
      requests:
        memory: "200Mi"
        cpu: "2"

The pod above runs in the Guaranteed QoS class because requests are equal to limits. And the container's resource limit for the CPU resource is an integer greater than or equal to one. The nginx container is granted 2 exclusive CPUs.

spec:
  containers:
  - name: nginx
    image: nginx
    resources:
      limits:
        memory: "200Mi"
        cpu: "1.5"
      requests:
        memory: "200Mi"
        cpu: "1.5"

The pod above runs in the Guaranteed QoS class because requests are equal to limits. But the container's resource limit for the CPU resource is a fraction. It runs in the shared pool.

spec:
  containers:
  - name: nginx
    image: nginx
    resources:
      limits:
        memory: "200Mi"
        cpu: "2"

The pod above runs in the Guaranteed QoS class because only limits are specified and requests are set equal to limits when not explicitly specified. And the container's resource limit for the CPU resource is an integer greater than or equal to one. The nginx container is granted 2 exclusive CPUs.

Static policy options

You can toggle groups of options on and off based upon their maturity level using the following feature gates:

• CPUManagerPolicyBetaOptions default enabled. Disable to hide beta-level options.
• CPUManagerPolicyAlphaOptions default disabled. Enable to show alpha-level options. You will still have to enable each option using the CPUManagerPolicyOptions kubelet option.

The following policy options exist for the static CPUManager policy:

• full-pcpus-only (beta, visible by default) (1.22 or higher)
• distribute-cpus-across-numa (alpha, hidden by default) (1.23 or higher)
• align-by-socket (alpha, hidden by default) (1.25 or higher)
• distribute-cpus-across-cores (alpha, hidden by default) (1.31 or higher)

If the full-pcpus-only policy option is specified, the static policy will always allocate full physical cores. By default, without this option, the static policy allocates CPUs using a topology-aware best-fit allocation. On SMT enabled systems, the policy can allocate individual virtual cores, which correspond to hardware threads. This can lead to different containers sharing the same physical cores; this behaviour in turn contributes to the noisy neighbours problem. With the option enabled, the pod will be admitted by the kubelet only if the CPU request of all its containers can be fulfilled by allocating full physical cores. If the pod does not pass the admission, it will be put in Failed state with the message SMTAlignmentError.

If the distribute-cpus-across-numapolicy option is specified, the static policy will evenly distribute CPUs across NUMA nodes in cases where more than one NUMA node is required to satisfy the allocation. By default, the CPUManager will pack CPUs onto one NUMA node until it is filled, with any remaining CPUs simply spilling over to the next NUMA node. This can cause undesired bottlenecks in parallel code relying on barriers (and similar synchronization primitives), as this type of code tends to run only as fast as its slowest worker (which is slowed down by the fact that fewer CPUs are available on at least one NUMA node). By distributing CPUs evenly across NUMA nodes, application developers can more easily ensure that no single worker suffers from NUMA effects more than any other, improving the overall performance of these types of applications.

If the align-by-socket policy option is specified, CPUs will be considered aligned at the socket boundary when deciding how to allocate CPUs to a container. By default, the CPUManager aligns CPU allocations at the NUMA boundary, which could result in performance degradation if CPUs need to be pulled from more than one NUMA node to satisfy the allocation. Although it tries to ensure that all CPUs are allocated from the minimum number of NUMA nodes, there is no guarantee that those NUMA nodes will be on the same socket. By directing the CPUManager to explicitly align CPUs at the socket boundary rather than the NUMA boundary, we are able to avoid such issues. Note, this policy option is not compatible with TopologyManager single-numa-node policy and does not apply to hardware where the number of sockets is greater than number of NUMA nodes.

If the distribute-cpus-across-cores policy option is specified, the static policy will attempt to allocate virtual cores (hardware threads) across different physical cores. By default, the CPUManager tends to pack cpus onto as few physical cores as possible, which can lead to contention among cpus on the same physical core and result in performance bottlenecks. By enabling the distribute-cpus-across-cores policy, the static policy ensures that cpus are distributed across as many physical cores as possible, reducing the contention on the same physical core and thereby improving overall performance. However, it is important to note that this strategy might be less effective when the system is heavily loaded. Under such conditions, the benefit of reducing contention diminishes. Conversely, default behavior can help in reducing inter-core communication overhead, potentially providing better performance under high load conditions.

The full-pcpus-only option can be enabled by adding full-pcpus-only=true to the CPUManager policy options. Likewise, the distribute-cpus-across-numa option can be enabled by adding distribute-cpus-across-numa=true to the CPUManager policy options. When both are set, they are "additive" in the sense that CPUs will be distributed across NUMA nodes in chunks of full-pcpus rather than individual cores. The align-by-socket policy option can be enabled by adding align-by-socket=true to the CPUManager policy options. It is also additive to the full-pcpus-only and distribute-cpus-across-numa policy options.

The distribute-cpus-across-cores option can be enabled by adding distribute-cpus-across-cores=true to the CPUManager policy options. It cannot be used with full-pcpus-only or distribute-cpus-across-numa policy options together at this moment.