AKS Cost Control: Cluster Stop, Tiers and Node Pools

Pricing tier first, nodes second

AKS offers three pricing tiers for cluster management. The Free tier has free cluster management, a best-effort uptime with no financially backed SLA, and Microsoft recommends it for development and test clusters with fewer than 10 nodes. Standard and Premium add an uptime SLA by default (99.95% for the API server with availability zones, 99.9% without), and Premium adds long-term support. AKS Automatic clusters are preconfigured on Standard. On every tier you pay as you go for the nodes and other resources the cluster consumes.

So the cluster fee depends on the tier you picked, and the node pools carry most of the bill regardless.

Stopping a whole cluster

Microsoft documents a cluster stop and start: stopping turns off the control plane and agent nodes, saving all the compute costs while keeping every object except standalone pods. The limits are worth reading before you schedule it:

• Only Virtual Machine Scale Set backed clusters are supported, and clusters using Node Autoprovisioning cannot be stopped.
• Stopped cluster state is preserved for up to 12 months.
• A stopped cluster accepts only start or delete; scaling or upgrading requires starting it first.
• In capacity-constrained regions a stopped cluster may fail to start, which is why Microsoft does not recommend stopping mission-critical workloads.

What ZopNight runs for AKS

AKS is on ZopNight’s start/stop list, and ZopNight applies the provider-specific operation for it. For finer control, AKS namespaces can be scheduled: Stop scales a namespace’s Deployments and StatefulSets to 0 and suspends CronJobs, saving replicas, HPA and PDB settings for Start to restore, and system namespaces are refused. Cluster start and stop run synchronously, without a background watcher.

Node pool recommendations from the Azure catalog: AKS Spot Node Pool Opportunity, AKS Node Pool Underutilized and AKS Agent Pool VM-Size Rightsizing (both of the latter propose a next-smaller same-family VM SKU priced from real rate data), plus AKS Node Pool Autoscaler Not Enabled. Cluster posture: AKS Cluster Network Policy Not Enabled, RBAC Not Enabled, Monitoring Not Enabled and API Server Not Private. Workload rules come from the shared Kubernetes framework under AKS IDs, including missing requests and limits, privileged containers and released PersistentVolumes. AKS activity comes from the Azure Activity Log, so idle workload rules see real Scale and Patch events rather than pod metrics alone.

Fact sheets: AKS clusters, agent pools. Schedules: AKS cluster, agent pool. More: Kubernetes workload rules.