Cluster Autoscaler

The Cluster Autoscaler automatically adjusts the number of nodes in the Kubernetes cluster based on pod scheduling demands. When pods cannot be scheduled due to insufficient resources, the Cluster Autoscaler adds nodes. When nodes are underutilized, it removes them to reduce costs.

Cluster Autoscaler Architecture

Unschedulable Pods --> Cluster Autoscaler --> Cloud Provider API --> Add Nodes
                            |
Underutilized Nodes --> Cluster Autoscaler --> Cloud Provider API --> Remove Nodes

Scaling Triggers

Scale Up

The Cluster Autoscaler adds nodes when:

Scale Down

The Cluster Autoscaler removes nodes when:

Node Pool Configuration

The MATIH platform uses multiple node pools for workload isolation:

Configuration Parameters

Cloud Provider Integration

Azure AKS Configuration

For the MATIH Azure deployment, Cluster Autoscaler is managed natively by AKS:

# Node pool autoscaling is configured via Terraform
resource "azurerm_kubernetes_cluster_node_pool" "dataplane" {
  name                = "dataplane"
  kubernetes_cluster_id = azurerm_kubernetes_cluster.main.id
  vm_size             = "Standard_D8s_v3"
  enable_auto_scaling = true
  min_count           = 2
  max_count           = 10
  node_labels = {
    "matih.io/node-pool" = "dataplane"
  }
}

Pod Disruption Budgets

Critical services have PDBs to prevent the autoscaler from removing nodes hosting essential pods:

Monitoring

Troubleshooting