GKE Cluster Autoscaler Location Policy: BALANCED vs ANY Explained

The GKE Cluster Autoscaler location policy controls how newly provisioned nodes are distributed across zones, with BALANCED evenly spreading nodes for high availability and ANY maximizing capacity acquisition from any available zone.

The GKE Cluster Autoscaler supports two distinct location policies that determine node distribution behavior during scale-up events. According to the google/skills repository documentation, these policies are configured via the --location-policy flag in gcloud or the locationPolicy field in a ComputeClass specification.

What Is the GKE Cluster Autoscaler Location Policy?

The location policy is a node-level configuration that instructs the Cluster Autoscaler how to select zones when provisioning new nodes. This decision happens at the infrastructure layer and affects only node placement, not pod scheduling.

As documented in ca-optimization.md, the autoscaler evaluates available capacity across zones and applies the selected policy to determine where new instances are created.

BALANCED Policy (Default)

The BALANCED policy directs the autoscaler to evenly distribute nodes across all available zones during scale-up operations. When the cluster needs additional capacity, the autoscaler attempts to maintain equilibrium in node counts between zones.

Use this policy for high-availability (HA) workloads where zone redundancy is critical. However, note that this node-level balancing does not influence pod placement. Applications requiring zone-distributed pods must implement additional constraints separately.

ANY Policy

The ANY policy allows the autoscaler to provision nodes from any zone that has available resources, without attempting to maintain balanced node counts across zones. The autoscaler simply selects the zone with the best available capacity at the moment of scaling.

This policy is optimal for Spot VMs or scarce machine SKUs where maximizing obtainability and minimizing provisioning latency takes priority over even distribution. As noted in ca-optimization.md, this prevents the autoscaler from waiting for capacity in under-provisioned zones.

How to Configure the Location Policy

You can configure the location policy using either the gcloud CLI or declarative YAML configurations via ComputeClass resources.

Using gcloud CLI

Update an existing cluster to use the balanced policy for even zone distribution:

gcloud container clusters update CLUSTER_NAME --location-policy=BALANCED

Switch to the any policy to optimize for Spot VM acquisition:

gcloud container clusters update CLUSTER_NAME --location-policy=ANY

These commands modify the cluster's autoscaling configuration immediately, as referenced in ca-provisioning.md.

Using ComputeClass YAML

For workload-specific configurations, define the location policy within a ComputeClass manifest. This approach allows different workloads to use different policies within the same cluster.

apiVersion: networking.gke.io/v1
kind: ComputeClass
metadata:
  name: spot-any
spec:
  priorities:
  - machineFamily: n4
    spot: true
    location:
      locationPolicy: ANY

In this example from the google/skills source, the ANY policy combined with spot: true ensures the autoscaler aggressively acquires Spot capacity from any available zone.

Location Policy vs Pod Topology Spread Constraints

A critical distinction exists between the autoscaler's location policy and pod-level spreading. The location policy controls node distribution, while pod distribution requires topologySpreadConstraints.

If your application requires pods to be balanced across zones, you must implement a Pod spec with topology spread constraints:

apiVersion: v1
kind: Pod
metadata:
  name: my-app
spec:
  topologySpreadConstraints:
  - maxSkew: 1
    topologyKey: "topology.kubernetes.io/zone"
    whenUnsatisfiable: DoNotSchedule
    labelSelector:
      matchLabels:
        app: my-app

The whenUnsatisfiable: DoNotSchedule setting is required for autoscaler compatibility, as documented in SKILL.md. Without this configuration, the scheduler may place multiple pods in one zone even when nodes exist in others, rendering the balanced node distribution ineffective for workload resilience.

Summary

• The BALANCED location policy evenly distributes nodes across zones for high-availability architectures, serving as the default behavior.
• The ANY location policy provisions nodes from any available zone to maximize capacity acquisition, ideal for Spot VMs and scarce resources.
• Configure policies via gcloud container clusters update --location-policy or the locationPolicy field in ComputeClass specifications.
• Node-level location policies do not affect pod placement; use topologySpreadConstraints with whenUnsatisfiable: DoNotSchedule for zone-balanced workloads.
• Reference implementation details in ca-optimization.md, ca-provisioning.md, and SKILL.md within the google/skills repository.

Frequently Asked Questions

Does the BALANCED location policy guarantee my pods are spread across zones?

No. The BALANCED policy only affects node distribution, not pod scheduling. The autoscaler attempts to create an equal number of nodes in each zone, but the Kubernetes scheduler may still place multiple pods on nodes in a single zone. To ensure pod distribution across zones, you must configure topologySpreadConstraints in your Pod spec with topologyKey: topology.kubernetes.io/zone.

When should I use the ANY location policy instead of BALANCED?

Use the ANY policy when running Spot VMs or workloads using scarce machine types where capacity availability varies significantly between zones. This policy minimizes provisioning time by allowing the autoscaler to immediately acquire resources from any zone with available capacity, rather than waiting to maintain balanced node counts. Avoid this policy for strict HA requirements that depend on node distribution symmetry.

How do I check which location policy my GKE cluster is currently using?

The location policy configuration is visible in your cluster's node pool autoscaling settings. Use the command gcloud container clusters describe CLUSTER_NAME and examine the autoscaling section. If no policy is explicitly set, the cluster uses BALANCED as the default. For ComputeClass-defined policies, inspect the specific ComputeClass resource using kubectl get computeclass CLASS_NAME -o yaml to verify the locationPolicy value.

Can different node pools in the same cluster use different location policies?

Yes. Location policies can be configured at the node pool level when using ComputeClass resources. By creating multiple ComputeClass manifests with different locationPolicy values and assigning them to specific workloads, you can run a BALANCED policy for standard HA workloads in one node pool while using ANY for Spot-based preemptible workloads in another. The cluster-level --location-policy flag serves as the default for node pools without specific ComputeClass overrides.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}