references/azure-aks-autoscaler.md
AKS Cluster Autoscaler (CAS)
Enable and tune the Cluster Autoscaler to automatically scale down idle nodes.
Check CAS Status
az aks show \
--name "<CLUSTER_NAME>" --resource-group "<RESOURCE_GROUP>" \
--query "agentPoolProfiles[].{name:name, casEnabled:enableAutoScaling, min:minCount, max:maxCount, count:count}" \
-o table
az aks show \
--name "<CLUSTER_NAME>" --resource-group "<RESOURCE_GROUP>" \
--query "autoScalerProfile" -o json
Check Node Utilization (7 days)
Follow the metrics discovery steps in azure-aks-rightsizing.md to list available metric definitions and query node CPU utilization. Use metric names such as node_cpu_usage_percentage or cpuUsagePercentage depending on what's available on the cluster.
Enable CAS
# Cluster-level
az aks update \
--name "<CLUSTER_NAME>" --resource-group "<RESOURCE_GROUP>" \
--enable-cluster-autoscaler \
--min-count <MIN_NODES> --max-count <MAX_NODES>
# Specific node pool
az aks nodepool update \
--cluster-name "<CLUSTER_NAME>" --resource-group "<RESOURCE_GROUP>" \
--name "<NODEPOOL_NAME>" \
--enable-cluster-autoscaler \
--min-count <MIN_NODES> --max-count <MAX_NODES>
Recommended min/max Defaults
| Scenario | min-count | max-count | |----------|-----------|-----------| | Dev/test | 1 | current_count | | Production (web/API) | 2 | current_count * 3 | | Production (batch) | 0 | current_count * 5 |
Risk: Low. CAS only scales down when pods can be safely rescheduled. Set min-count >= 2 for production HA.
Tune CAS Profile
Apply when CAS is already on but idle nodes persist. The following cost-oriented profile is an example, not a universal default. The two scale-down timers are shorter than their documented 10-minute defaults, while the utilization threshold remains at its 0.5 default:
az aks update \
--name "<CLUSTER_NAME>" --resource-group "<RESOURCE_GROUP>" \
--cluster-autoscaler-profile \
scale-down-delay-after-add=5m \
scale-down-unneeded-time=5m \
scale-down-utilization-threshold=0.5 \
max-graceful-termination-sec=600
Risk: Medium. Shorter scale-down timers can reduce idle cost, but they can also cause node reprovisioning after brief demand changes. This thrashing adds provisioning latency; use a less aggressive profile for workloads with frequent short scale-out and scale-in cycles. See AKS cluster autoscaler profile settings and aggressive scale-down guidance.
Keep
skip-nodes-with-system-podsat its defaulttrue. It prevents scale-down of nodes that host non-DaemonSetkube-systempods. Do not useskip-nodes-with-system-pods=falseas a cost lever or assume every Microsoft-managed system component can be protected by a customer-managed PDB. See AKS support policies.
To roll back to CAS defaults:
az aks update \
--name "<CLUSTER_NAME>" --resource-group "<RESOURCE_GROUP>" \
--cluster-autoscaler-profile ""
Profile Comparison
| Profile | scale-down-delay-after-add | scale-down-unneeded-time | utilization-threshold | Best For | |---------|----------------------------|--------------------------|----------------------|----------| | Default | 10m | 10m | 0.5 | General workloads | | Cost-Optimized | 5m | 5m | 0.5 | Cost-sensitive, non-critical | | Conservative | 30m | 30m | 0.7 | Stateful / production | | Aggressive | 2m | 2m | 0.4 | Dev/test, batch |
Risk: High for aggressive tuning. Ensure PodDisruptionBudgets (PDBs) are set on critical workloads before tuning. Always confirm with user before applying.
Check existing PDBs before tuning:
kubectl get pdb --all-namespaces