references/aks-savings-plans-reservations.md
AKS Commitment-Demand Handoff
Prepare stable, post-optimization AKS compute demand for cost-optimization,
which owns Reservation and Savings Plan analysis. Direct commitment requests
start there. Use this workflow only when AKS optimization could materially
change steady node demand; do not calculate economics or purchase commitments.
Preconditions
Complete applicable rightsizing, autoscaler, NAP, and bin-packing analysis. Confirm the cluster, representative lookback, planned compute changes, and retained Capacity Reservation Group (CRG) requirements.
Collect AKS Demand
- Use Azure MCP AKS or an approved
az aks showfallback for node resource group, region, provisioning mode, and regular pool metadata. - For manual provisioning, collect each pool's VM size, current/min/max count, autoscaling, priority, CRG association, and upgrade surge. Map it to exact VMSS or VM resource IDs from pool metadata, not names alone.
- For NAP, map each observed node or
NodeClaimthroughproviderIDto exact Azure compute IDs; do not infer demand from the AKS pool list. Retain time-aligned SKU, region, zone, architecture, OS, andkarpenter.sh/capacity-type; report unmapped resources as an evidence gap. - Exclude Spot capacity from stable commitment demand. Group non-Spot observations by compatible compute characteristics.
- Reuse representative history; otherwise range-query managed Prometheus
kube_node_infohourly over the requested 30-60 days, joining node andprovider_idto the time-aligned mapping. Fall back to Container Insights only when Prometheus history is unsuitable. Report the requested/observed periods, observed hours, min/p05/p50/p95/max counts, and candidate-baseline hours and coverage. Retention limits the lookback. If neither source provides representative history, stable demand is unknown and no baseline may be produced. - Query associated CRGs through Resource Graph. Keep SKU, region, zone, reserved, associated, unused, and retained quantities separate. Unused CRG capacity remains billable.
- Account for approved rightsizing, consolidation, NAP, migration, and reliability changes. Record seasonal, temporary, surge, outage, and planned demand that makes the lookback unrepresentative.
Handoff
Provide cost-optimization with:
| Field | Content | |---|---| | AKS scope | Cluster ID, node resource group, region | | Compute mapping | Provisioning mode, compatible compute group, exact VMSS/VM resource IDs, SKU, region/zone, architecture, OS, capacity type | | Continuous baseline | Candidate non-Spot counts by compatible compute group, hours covered, coverage percentage, p05/p50/p95, lookback | | Variability | Peak, seasonal, temporary, surge, outage, and planned-demand exceptions | | Planned changes | Rightsizing, consolidation, NAP, migration, SKU/region changes | | Capacity reservations | CRG reserved, associated, unused, and retained quantities | | Evidence gaps | Missing history, inaccessible pools, or unresolved owner intent |
Ask cost-optimization to evaluate utilization, coverage, scope, expiry, and
renewal; retrieve Azure recommendations; correlate them with exact AKS mappings
and approved changes; disclose broader-scope benefits; and return term, scope,
coverage, expected utilization, reported savings/currency, assumptions, risks,
and next review date.
This workflow does not query benefit APIs, recommend quantities, or claim
expiry/renewal actions. If cost-optimization is unavailable, return the
demand evidence and mark commitment analysis unavailable.