How to configure scheduling for ACK Serverless clusters-Container Service for Kubernetes(ACK)-阿里云帮助中心

Pods default to Linux/x86 nodes; configure scheduling to target other architectures or zones.

In ACK Serverless clusters, all pods are Elastic Container Instance (ECI)-based pods that run on virtual nodes. ACK Serverless supports three Kubernetes scheduling primitives: nodeSelector, nodeAffinity/podAffinity, and topologySpreadConstraints.

Virtual node types

ACK Serverless clusters provide three virtual node types:

Node type	CPU architecture	Operating system
Default virtual nodes	x86	Linux
ARM-based virtual nodes	ARM	Linux
Windows virtual nodes	—	Windows

Choose a scheduling method

Each primitive offers a different constraint level and fallback behavior.

Method	Use when	Constraint type
`nodeSelector`	Target a specific CPU architecture or OS	Hard (required)
`nodeAffinity` and `podAffinity`	Prefer a zone or architecture, or co-locate pods — with fallback options	Hard or soft (preferred)
`topologySpreadConstraints`	Spread pods evenly across zones for high availability	Distribution control

Key distinction: nodeSelector is a hard requirement — pods fail to schedule without a match. nodeAffinity with preferredDuringSchedulingIgnoredDuringExecution is a soft preference — pods still schedule without a preferred match. Use soft affinity for multi-architecture images that prefer ARM.

Note

ACK Serverless Basic clusters support only nodeSelector, and only for the kubernetes.io/arch and kubernetes.io/os labels. Zone affinity, pod affinity, and topology spreading require an ACK Serverless Pro cluster, which provides enhanced reliability, higher SLA guarantees, and larger cluster capacity. See Hot migration from ACK Serverless Basic to Pro.

Cluster scheduling capabilities

Scheduling method	ACK Serverless Basic	ACK Serverless Pro
`nodeSelector` (`kubernetes.io/arch`, `kubernetes.io/os`)	Supported	Supported
`nodeAffinity`	Not supported	Supported
`podAffinity`	Not supported	Supported
`topologySpreadConstraints`	Not supported	Supported

Scheduling methods

nodeSelector

Add a nodeSelector field to target virtual nodes with a specific CPU architecture or OS.

spec:
  nodeSelector:
    kubernetes.io/arch: arm64   # Target ARM-based virtual nodes
    # kubernetes.io/os: windows # Uncomment to target Windows virtual nodes

References:

nodeAffinity and podAffinity

Use nodeAffinity to constrain pods to specific zones or preferred architectures. Use podAffinity to co-locate related pods.

This example uses soft node affinity to prefer ARM nodes for a multi-architecture workload:

spec:
  affinity:
    nodeAffinity:
      preferredDuringSchedulingIgnoredDuringExecution:
        - weight: 1
          preference:
            matchExpressions:
              - key: kubernetes.io/arch
                operator: In
                values:
                  - arm64

References:

topologySpreadConstraints

Spread pods across zones for high availability. maxSkew controls the maximum allowed imbalance between zones.

spec:
  topologySpreadConstraints:
    - maxSkew: 1
      topologyKey: topology.kubernetes.io/zone
      whenUnsatisfiable: DoNotSchedule
      labelSelector:
        matchLabels:
          app: my-app

With maxSkew: 1, no zone has more than one extra pod compared to the least-loaded zone.

References: