Help Center/ Cloud Container Engine/ User Guide/ Workloads/ Scheduling a Workload/ Configuring Workload Affinity or Anti-affinity Scheduling (podAffinity or podAntiAffinity)
Updated on 2024-11-12 GMT+08:00

Configuring Workload Affinity or Anti-affinity Scheduling (podAffinity or podAntiAffinity)

Kubernetes offers workload affinity and anti-affinity scheduling, which allows for flexible scheduling of new workloads on either related or unrelated nodes. This results in improved cluster resource utilization.

For example, frontend workload pods and backend workload pods that frequently communicate with each other can be preferentially scheduled to the same node or AZ to minimize network latency. The process of workload affinity/anti-affinity is as follows:

  1. Nodes are classified using node labels based on topology keys (topologyKey).

    For example, if topologyKey is prefer, nodes will be classified using node label prefer. Nodes labeled with prefer=true will be assigned to topology key 1, while those labeled with prefer=false will be assigned to topology key 2. Nodes without a prefer label will be assigned to topology key 3.

  2. Identify the affinity/anti-affinity workloads based on the workload labels and operators.

    For example, the label selector filters the workloads that have been labeled with app=backend.

  3. For affinity scheduling, the scheduler chooses the topology key where the target workload is located, while for anti-affinity scheduling, it selects a topology key where the target workload is not present.

    In this example, the workload labeled with app=backend is assigned to topology key 1. Therefore, workloads that have an affinity with the app=backend workload can be scheduled to topology key 1. Conversely, workloads that have an anti-affinity with the app=backend workload can only be scheduled to topology key 2 or 3.

Figure 1 Workload affinity or anti-affinity scheduling

Configuring Load Affinity/Anti-affinity

There are various ways to configure load affinity or anti-affinity scheduling policies to schedule pods to nodes that meet specific requirements.

In this example, a backend workload with the app=backend label has been created in the cluster. This label can be used for workload affinity or anti-affinity scheduling, allowing the newly created frontend workload labeled with app=frontend to be deployed on the same node as the backend workload. Both workloads run in topology key kubernetes.io/hostname. When you use kubernetes.io/hostname for topology classification, only one node can be assigned to each topology key because each node has a unique hostname. This enables workloads to be scheduled on the same node when workload affinity is enabled.

  1. Log in to the CCE console.
  2. Click the cluster name to go to the cluster console, choose Workloads in the navigation pane, and click the Create Workload in the upper right corner.
  3. In Advanced Settings, choose Scheduling and select a policy for Load Affinity. In this example, Custom policies is selected. For details about how to create a workload, see Creating a Workload.

    Table 1 Scheduling policies

    Parameter

    Description

    Example

    Load affinity

    • Not configured: No load affinity policy is configured.
    • Multi-AZ deployment preferred: Workload pods are preferentially scheduled to nodes in different AZs through pod anti-affinity. The AZs serve as topology keys in this process.
    • Forcible multi-AZ deployment: Workload pods are forcibly scheduled to nodes in different AZs through pod anti-affinity. The AZs serve as topology keys in this process. When this scheduling policy is used, if there are fewer nodes than pods or node resources are insufficient, the extra pods will fail to run.
    • Custom policies: allow flexible scheduling of workload pods based on pod labels. For details about the supported scheduling policies, see Table 2. Select a proper policy type and add a policy. For details about the parameters, see Table 3.

    Custom policies

  4. Select a proper load affinity rule and click to add a scheduling policy. In this example, a scheduling policy is added in the Required area under Workload Affinity. This policy specifies that the newly created workload can only be scheduled to a node if a workload with specific labels is already running on that node.

    Table 2 Load affinity policies

    Policy

    Rule Type

    Description

    Example

    Workload affinity

    Required

    Hard constraint, which corresponds to requiredDuringSchedulingIgnoredDuringExecution in YAML for specifying the conditions that must be met.

    Select pods that require affinity by label. If such pods already run on a node in the topology key, the scheduler will forcibly schedule the created pods to that topology key.

    NOTE:

    If multiple affinity rules are configured, multiple labels will be used to filter pods that require affinity, and the newly created pods must be affinity with all pods that meet the label filtering conditions. In this way, all pods that meet the label filtering conditions locate in the same topology key for scheduling.

    Required

    Preferred

    Soft constraint, which corresponds to preferredDuringSchedulingIgnoredDuringExecution in YAML for specifying the conditions that need to be met as much as possible.

    Select pods that require affinity by label. If such pods already run on a node in the topology key, the scheduler will preferentially schedule the created pods to that topology key.

    NOTE:

    If multiple affinity rules are configured, multiple labels will be used to filter pods that require affinity, and the newly created pods will be preferentially to be affinity with multiple pods that meet the label filtering conditions. However, even if no pod meets the label filter conditions, a topology key will be selected for scheduling.

    Workload anti-affinity

    Required

    Hard constraint, which corresponds to requiredDuringSchedulingIgnoredDuringExecution in YAML for specifying the conditions that must be met.

    Select one or more pods that require anti-affinity by label. If such pods already run on a node in the topology key, the scheduler will not schedule the created pods to that topology key.

    NOTE:

    If multiple anti-affinity rules are configured, multiple labels will be used to filter pods that require anti-affinity, and the newly created pods must be anti-affinity with all pods that meet the label filtering conditions. In this way, all the topology keys where the pods that meet the label filtering conditions locate will not be scheduled.

    None

    Preferred

    Soft constraint, which corresponds to preferredDuringSchedulingIgnoredDuringExecution in YAML for specifying the conditions that need to be met as much as possible.

    Select one or more pods that require anti-affinity by label. If such pods already run on a node in the topology key, the scheduler will preferentially schedule the created pods to other topology keys.

    NOTE:

    If multiple anti-affinity rules are configured, multiple labels will be used to filter pods that require anti-affinity, and the newly created pods will be preferentially to be anti-affinity with multiple pods that meet the label filtering conditions. However, even if all topology keys involve the pods that require anti-affinity, a topology key will be selected for scheduling.

  5. In the dialog box that is displayed on the right, click Add Policy to configure rules for filtering node labels.

    Table 3 Parameters for configuring load affinity/anti-affinity scheduling policies

    Parameter

    Description

    Example

    Weight

    This parameter is available only in a preferred scheduling policy. Weights range from 1 to 100 and are taken into account as an extra scoring factor during scheduling. The scheduler combines the weight with other priority functions of the node to determine the final score and then assigns pods to the node with the highest total score.

    None

    Namespace

    Namespace for which the scheduling policy takes effect.

    default

    Topology Key

    A topology key (topologyKey) determines the range of nodes to be scheduled based on node labels, identifies affinity/anti-affinity objects based on the labels and operators, and performs scheduling based on the topology key where the target object is located.

    • For example, if the node label is kubernetes.io/hostname, the label value will be a node name. Nodes with different names are assigned to different topology keys. This allows for workload affinity scheduling on a single node, as each topology key contains only one node.
    • If the specified label is kubernetes.io/os, the label value will be a node OS. Nodes running different OSs are assigned to different topology keys. This allows for workload affinity scheduling on multiple nodes, as each topology key contains multiple nodes.

      For example, if pods that meet the load affinity rule are running on a node in a topology key, all nodes in the topology key can be scheduled.

    kubernetes.io/hostname

    Label Key

    When configuring a workload affinity or anti-affinity policy, enter the workload label to be matched.

    Both default labels and custom labels are supported.

    app

    Operator

    The following operators are supported:

    • In: The label of the affinity or anti-affinity object is in the label value list (values field).
    • NotIn: The label of the affinity or anti-affinity object is not in the label value list (values field).
    • Exists: The affinity or anti-affinity object has a specified label key.
    • DoesNotExist: The affinity or anti-affinity object does not have a specified label key.

    In

    Label Value

    When configuring a workload affinity or anti-affinity policy, enter the value of the workload label.

    backend

  6. After the scheduling policy is added, click Create Workload.
  7. Verify that all pods run on the target node.

    1. In the navigation pane, choose Workloads.
    2. Click the workload name to enter its details page. On the Pods tab page, verify that the new pod and the existing backend pod run on the same node.

  • Workload affinity

    Kubernetes supports affinity between pods, which allows the frontend and backend pods of an application to be deployed together to minimize access latency.

    Assume that the backend pods of an application have been created with label app=backend. You can use .spec.affinity.podAffinity to configure workload affinity so that the frontend pods (labeled app=frontend) and backend pods (labeled app=backend) can be deployed together.

    apiVersion: apps/v1
    kind: Deployment
    metadata:
      name: frontend
      labels:
        app: frontend
    spec:
      selector:
        matchLabels:
          app: frontend
      replicas: 3
      template:
        metadata:
          labels:
            app: frontend
        spec:
          containers:
          - image: nginx:alpine
            name: frontend
            resources:
              requests:
                cpu: 100m
                memory: 200Mi
              limits:
                cpu: 100m
                memory: 200Mi
          imagePullSecrets:
          - name: default-secret
          affinity:  # Configure a scheduling policy.
            podAffinity:  # Workload affinity scheduling rule
              requiredDuringSchedulingIgnoredDuringExecution:   # Scheduling policy that must be met
              - topologyKey: prefer    # Topology keys are divided based on node labels, among which prefer is a custom label.
                labelSelector:  # Select workloads that meet the requirements based on workload labels.
                  matchExpressions: # Workload label matching rule
                  - key: app # The key of the workload label is app.
                    operator: In # The rule is met if a value exists in the value list.
                    values: # Workload label values
                    - backend
              preferredDuringSchedulingIgnoredDuringExecution:    # Scheduling policy that is met as much as possible
              - weight: 100  # Priority that can be configured when the best-effort policy is used. The value ranges from 1 to 100. A larger value indicates a higher priority.
                podAffinityTerm:  # Affinity configuration when the best-effort policy is used
                  topologyKey: topology.kubernetes.io/zone   # Topology keys are divided based on node labels by node AZ.
                  labelSelector:
                    matchExpressions:
                    - key: app
                      operator: In
                      values:
                      - backend

    During workload scheduling in the preceding example, node topology keys are divided based on the prefer label using the rule that must be met. If backend pods (labeled app=backend) are running on a node in the topology key, frontend pods (labeled app=frontend) will also be deployed in that topology key, even if not all nodes in the topology key are running the backend pods. According to the best-effort rule, topology keys are divided based on topology.kubernetes.io/zone by node AZ. This ensures that the frontend and backend pods are deployed on nodes within the same AZ as much as possible.

    For workload affinity, topologyKey cannot be left blank when requiredDuringSchedulingIgnoredDuringExecution and preferredDuringSchedulingIgnoredDuringExecution are used.

    topologyKey is used to divide topology keys based on the labels of nodes. Nodes with the same labels are grouped into the same topology key. The scheduler then selects the topology key to be scheduled based on the workload label. A topology key can consist of multiple nodes. If a workload that meets a label selection rule runs on a node in a topology key, all nodes in the topology key can be scheduled.

    For example, if the topologyKey label is set to topology.kubernetes.io/zone, nodes' AZs will be used as the topology keys, and workloads will be scheduled by AZ during deployment.

  • Workload anti-affinity

    In some cases, pods need to be deployed separately. This is because deploying them together can negatively impact performance.

    Assume that the frontend pods of an application have been created with label app=frontend. To ensure that pods are deployed on different nodes and multiple AZs are preferred, you can use .spec.affinity.podAntiAffinity to configure workload anti-affinity.

    apiVersion: apps/v1
    kind: Deployment
    metadata:
      name:   frontend
      labels:
        app:  frontend
    spec:
      selector:
        matchLabels:
          app: frontend
      replicas: 5
      template:
        metadata:
          labels:
            app:  frontend
        spec:
          containers:
          - image:  nginx:alpine
            name:  frontend
            resources:
              requests:
                cpu:  100m
                memory:  200Mi
              limits:
                cpu:  100m
                memory:  200Mi
          imagePullSecrets:
          - name: default-secret
          affinity:
            podAntiAffinity:  # Workload anti-affinity scheduling rule
              requiredDuringSchedulingIgnoredDuringExecution:   # Scheduling policy that must be met
              - topologyKey: kubernetes.io/hostname    # Topology keys are divided based on node labels.
                labelSelector:    # Pod label matching rule
                  matchExpressions:  # The key of the workload label is app.
                  - key: app  # The key of the workload label is app.
                    operator: In  # The rule is met if a value exists in the value list.
                    values:  # Workload label values
                    - frontend
              preferredDuringSchedulingIgnoredDuringExecution:    # Scheduling policy that is met as much as possible
              - weight: 100  # Priority that can be configured when the best-effort policy is used. The value ranges from 1 to 100. A larger value indicates a higher priority.
                podAffinityTerm:  # Affinity configuration when the best-effort policy is used
                  topologyKey: topology.kubernetes.io/zone   # Topology keys are divided based on node labels.
                  labelSelector:
                    matchExpressions:
                    - key: app
                      operator: In
                      values:
                      - frontend

    In the preceding example, anti-affinity rules are configured. The rule that must be met indicates that node topology keys are divided based on kubernetes.io/hostname. Nodes with the kubernetes.io/hostname label have different label values. Therefore, there is only one node in each topology key. If a topology key contains only one node where a frontend pod already exists, pods with the same label will not be scheduled to that topology key. According to the best-effort rule, topology keys are divided based on topology.kubernetes.io/zone by node AZ. This ensures that the pods are deployed on nodes in different AZs as much as possible.

    For workload anti-affinity, when requiredDuringSchedulingIgnoredDuringExecution is used, the default access controller LimitPodHardAntiAffinityTopology of Kubernetes requires that topologyKey can only be kubernetes.io/hostname. To use other custom topology logic, modify or disable the access controller.