Chapter 4. Controlling pod placement onto nodes (scheduling)

The default OpenShift Dedicated pod scheduler is responsible for determining the placement of new pods onto nodes within the cluster. It reads data from the pod and finds a node that is a good fit based on configured profiles. It is completely independent and exists as a standalone solution. It does not modify the pod; it creates a binding for the pod that ties the pod to the particular node.

One of the important use cases for scheduling within OpenShift Dedicated is to support flexible affinity and anti-affinity policies.

Pod affinity and pod anti-affinity allow you to constrain which nodes your pod is eligible to be scheduled on based on the key/value labels on other pods.

For example, using affinity rules, you could spread or pack pods within a service or relative to pods in other services. Anti-affinity rules allow you to prevent pods of a particular service from scheduling on the same nodes as pods of another service that are known to interfere with the performance of the pods of the first service. Or, you could spread the pods of a service across nodes, availability zones, or availability sets to reduce correlated failures.

There are two types of pod affinity rules: required and preferred.

Required rules must be met before a pod can be scheduled on a node. Preferred rules specify that, if the rule is met, the scheduler tries to enforce the rules, but does not guarantee enforcement.

You configure pod affinity/anti-affinity through the Pod spec files. You can specify a required rule, a preferred rule, or both. If you specify both, the node must first meet the required rule, then attempts to meet the preferred rule.

The following example shows a Pod spec configured for pod affinity and anti-affinity.

In this example, the pod affinity rule indicates that the pod can schedule onto a node only if that node has at least one already-running pod with a label that has the key security and value S1. The pod anti-affinity rule says that the pod prefers to not schedule onto a node if that node is already running a pod with label having key security and value S2.

apiVersion: v1
kind: Pod
metadata:
  name: with-pod-affinity
spec:
  securityContext:
    runAsNonRoot: true
    seccompProfile:
      type: RuntimeDefault
  affinity:
    podAffinity: 1
      requiredDuringSchedulingIgnoredDuringExecution: 2
      - labelSelector:
          matchExpressions:
          - key: security 3
            operator: In 4
            values:
            - S1 5
        topologyKey: topology.kubernetes.io/zone
  containers:
  - name: with-pod-affinity
    image: docker.io/ocpqe/hello-pod
    securityContext:
      allowPrivilegeEscalation: false
      capabilities:
        drop: [ALL]

apiVersion: v1
kind: Pod
metadata:
  name: with-pod-antiaffinity
spec:
  securityContext:
    runAsNonRoot: true
    seccompProfile:
      type: RuntimeDefault
  affinity:
    podAntiAffinity: 1
      preferredDuringSchedulingIgnoredDuringExecution: 2
      - weight: 100  3
        podAffinityTerm:
          labelSelector:
            matchExpressions:
            - key: security 4
              operator: In 5
              values:
              - S2
          topologyKey: kubernetes.io/hostname
  containers:
  - name: with-pod-affinity
    image: docker.io/ocpqe/hello-pod
    securityContext:
      allowPrivilegeEscalation: false
      capabilities:
        drop: [ALL]

The following steps demonstrate a simple two-pod configuration that creates pod with a label and a pod that uses affinity to allow scheduling with that pod.

The following steps demonstrate a simple two-pod configuration that creates pod with a label and a pod that uses an anti-affinity preferred rule to attempt to prevent scheduling with that pod.

The following examples demonstrate pod affinity and pod anti-affinity.

Node affinity allows a pod to specify an affinity towards a group of nodes it can be placed on. The node does not have control over the placement.

For example, you could configure a pod to only run on a node with a specific CPU or in a specific availability zone.

There are two types of node affinity rules: required and preferred.

Required rules must be met before a pod can be scheduled on a node. Preferred rules specify that, if the rule is met, the scheduler tries to enforce the rules, but does not guarantee enforcement.

You configure node affinity through the Pod spec file. You can specify a required rule, a preferred rule, or both. If you specify both, the node must first meet the required rule, then attempts to meet the preferred rule.

The following example is a Pod spec with a rule that requires the pod be placed on a node with a label whose key is e2e-az-NorthSouth and whose value is either e2e-az-North or e2e-az-South:

apiVersion: v1
kind: Pod
metadata:
  name: with-node-affinity
spec:
  securityContext:
    runAsNonRoot: true
    seccompProfile:
      type: RuntimeDefault
  affinity:
    nodeAffinity: 1
      requiredDuringSchedulingIgnoredDuringExecution: 2
        nodeSelectorTerms:
        - matchExpressions:
          - key: e2e-az-NorthSouth 3
            operator: In 4
            values:
            - e2e-az-North 5
            - e2e-az-South 6
  containers:
  - name: with-node-affinity
    image: docker.io/ocpqe/hello-pod
    securityContext:
      allowPrivilegeEscalation: false
      capabilities:
        drop: [ALL]
# ...

The following example is a node specification with a preferred rule that a node with a label whose key is e2e-az-EastWest and whose value is either e2e-az-East or e2e-az-West is preferred for the pod:

apiVersion: v1
kind: Pod
metadata:
  name: with-node-affinity
spec:
  securityContext:
    runAsNonRoot: true
    seccompProfile:
      type: RuntimeDefault
  affinity:
    nodeAffinity: 1
      preferredDuringSchedulingIgnoredDuringExecution: 2
      - weight: 1 3
        preference:
          matchExpressions:
          - key: e2e-az-EastWest 4
            operator: In 5
            values:
            - e2e-az-East 6
            - e2e-az-West 7
  containers:
  - name: with-node-affinity
    image: docker.io/ocpqe/hello-pod
    securityContext:
      allowPrivilegeEscalation: false
      capabilities:
        drop: [ALL]
# ...

There is no explicit node anti-affinity concept, but using the NotIn or DoesNotExist operator replicates that behavior.

Required rules must be met before a pod can be scheduled on a node.

Preferred rules specify that, if the rule is met, the scheduler tries to enforce the rules, but does not guarantee enforcement.

The following examples demonstrate node affinity.

Requests and limits enable administrators to allow and manage the overcommitment of resources on a node. The scheduler uses requests for scheduling your container and providing a minimum service guarantee. Limits constrain the amount of compute resource that may be consumed on your node.

OpenShift Dedicated administrators can control the level of overcommit and manage container density on nodes by configuring masters to override the ratio between request and limit set on developer containers. In conjunction with a per-project LimitRange object specifying limits and defaults, this adjusts the container limit and request to achieve the desired level of overcommit.

After these overrides, the container limits and requests must still be validated by any LimitRange object in the project. It is possible, for example, for developers to specify a limit close to the minimum limit, and have the request then be overridden below the minimum limit, causing the pod to be forbidden. This unfortunate user experience should be addressed with future work, but for now, configure this capability and LimitRange objects with caution.

In an overcommitted environment, it is important to properly configure your node to provide best system behavior.

When the node starts, it ensures that the kernel tunable flags for memory management are set properly. The kernel should never fail memory allocations unless it runs out of physical memory.

To ensure this behavior, OpenShift Dedicated configures the kernel to always overcommit memory by setting the vm.overcommit_memory parameter to 1, overriding the default operating system setting.

OpenShift Dedicated also configures the kernel not to panic when it runs out of memory by setting the vm.panic_on_oom parameter to 0. A setting of 0 instructs the kernel to call oom_killer in an Out of Memory (OOM) condition, which kills processes based on priority.

You can view the current setting by running the following commands on your nodes:

$ sysctl -a |grep commit

#...
vm.overcommit_memory = 0
#...

$ sysctl -a |grep panic

#...
vm.panic_on_oom = 0
#...

You can also perform the following configurations for each node:

You can use node selectors on pods and labels on nodes to control where the pod is scheduled. With node selectors, OpenShift Dedicated schedules the pods on nodes that contain matching labels.

You can use a node selector to place specific pods on specific nodes, cluster-wide node selectors to place new pods on specific nodes anywhere in the cluster, and project node selectors to place new pods in a project on specific nodes.

For example, as a cluster administrator, you can create an infrastructure where application developers can deploy pods only onto the nodes closest to their geographical location by including a node selector in every pod they create. In this example, the cluster consists of five data centers spread across two regions. In the U.S., label the nodes as us-east, us-central, or us-west. In the Asia-Pacific region (APAC), label the nodes as apac-east or apac-west. The developers can add a node selector to the pods they create to ensure the pods get scheduled on those nodes.

A pod is not scheduled if the Pod object contains a node selector, but no node has a matching label.

You can use node selectors on pods and labels on nodes to control where the pod is scheduled. With node selectors, OpenShift Dedicated schedules the pods on nodes that contain matching labels.

You add labels to a node, a compute machine set, or a machine config. Adding the label to the compute machine set ensures that if the node or machine goes down, new nodes have the label. Labels added to a node or machine config do not persist if the node or machine goes down.

To add node selectors to an existing pod, add a node selector to the controlling object for that pod, such as a ReplicaSet object, DaemonSet object, StatefulSet object, Deployment object, or DeploymentConfig object. Any existing pods under that controlling object are recreated on a node with a matching label. If you are creating a new pod, you can add the node selector directly to the pod spec. If the pod does not have a controlling object, you must delete the pod, edit the pod spec, and recreate the pod.

$ oc describe pod router-default-66d5cf9464-7pwkc

kind: Pod
apiVersion: v1
metadata:
# ...
Name:               router-default-66d5cf9464-7pwkc
Namespace:          openshift-ingress
# ...
Controlled By:      ReplicaSet/router-default-66d5cf9464
# ...

The web console lists the controlling object under ownerReferences in the pod YAML:

apiVersion: v1
kind: Pod
metadata:
  name: router-default-66d5cf9464-7pwkc
# ...
  ownerReferences:
    - apiVersion: apps/v1
      kind: ReplicaSet
      name: router-default-66d5cf9464
      uid: d81dd094-da26-11e9-a48a-128e7edf0312
      controller: true
      blockOwnerDeletion: true
# ...

Skew refers to the difference in the number of pods that match a specified label selector across different topology domains, such as zones or nodes.

The skew is calculated for each domain by taking the absolute difference between the number of pods in that domain and the number of pods in the domain with the lowest amount of pods scheduled. Setting a maxSkew value guides the scheduler to maintain a balanced pod distribution.

You can specify which pods to group together, which topology domains they are spread among, and the acceptable skew.

The following examples demonstrate pod topology spread constraint configurations.

apiVersion: v1
kind: Pod
metadata:
  name: my-pod
  labels:
    region: us-east
spec:
  securityContext:
    runAsNonRoot: true
    seccompProfile:
      type: RuntimeDefault
  topologySpreadConstraints:
  - maxSkew: 1 1
    topologyKey: topology.kubernetes.io/zone 2
    whenUnsatisfiable: DoNotSchedule 3
    labelSelector: 4
      matchLabels:
        region: us-east 5
    matchLabelKeys:
      - my-pod-label 6
  containers:
  - image: "docker.io/ocpqe/hello-pod"
    name: hello-pod
    securityContext:
      allowPrivilegeEscalation: false
      capabilities:
        drop: [ALL]

kind: Pod
apiVersion: v1
metadata:
  name: my-pod
  labels:
    region: us-east
spec:
  securityContext:
    runAsNonRoot: true
    seccompProfile:
      type: RuntimeDefault
  topologySpreadConstraints:
  - maxSkew: 1
    topologyKey: topology.kubernetes.io/zone
    whenUnsatisfiable: DoNotSchedule
    labelSelector:
      matchLabels:
        region: us-east
  containers:
  - image: "docker.io/ocpqe/hello-pod"
    name: hello-pod
    securityContext:
      allowPrivilegeEscalation: false
      capabilities:
        drop: [ALL]

The previous example defines a Pod spec with a one pod topology spread constraint. It matches on pods labeled region: us-east, distributes among zones, specifies a skew of 1, and does not schedule the pod if it does not meet these requirements.

kind: Pod
apiVersion: v1
metadata:
  name: my-pod-2
  labels:
    region: us-east
spec:
  securityContext:
    runAsNonRoot: true
    seccompProfile:
      type: RuntimeDefault
  topologySpreadConstraints:
  - maxSkew: 1
    topologyKey: node
    whenUnsatisfiable: DoNotSchedule
    labelSelector:
      matchLabels:
        region: us-east
  - maxSkew: 1
    topologyKey: rack
    whenUnsatisfiable: DoNotSchedule
    labelSelector:
      matchLabels:
        region: us-east
  containers:
  - image: "docker.io/ocpqe/hello-pod"
    name: hello-pod
    securityContext:
      allowPrivilegeEscalation: false
      capabilities:
        drop: [ALL]

The previous example defines a Pod spec with two pod topology spread constraints. Both match on pods labeled region: us-east, specify a skew of 1, and do not schedule the pod if it does not meet these requirements.

The first constraint distributes pods based on a user-defined label node, and the second constraint distributes pods based on a user-defined label rack. Both constraints must be met for the pod to be scheduled.