references/nap-enablement-and-rollback.md
NAP Enablement and Rollback
Use this procedure only after NAP readiness produces a recommendation and the user requests enablement.
Approve the Change
Resolve every eligibility and readiness gap identified during the NAP
assessment before enablement. Obtain approval for the default configuration or
proposed custom NodePool/AKSNodeClass settings. Explain capacity limits,
disruption behavior, including proposed expiration or termination deadlines,
and any CAS migration.
Before enablement, agree on an observation period, cost and capacity bounds, workload SLOs, success and failure criteria, and rollback. Verify that fixed pools can be created or scaled to absorb all NAP workloads during rollback, including DaemonSet overhead and operational headroom.
Enable and Verify
For default pools with CAS disabled:
az aks update \
--name "<CLUSTER_NAME>" --resource-group "<RESOURCE_GROUP>" \
--node-provisioning-mode Auto
For custom pools, add --node-provisioning-default-pools None to disable
default-pool creation, then apply the approved NodePool and AKSNodeClass
manifests.
Verify pending-pod time, provisioned configurations, node count, workload SLOs, and cost against the agreed criteria. Do not remove fallback fixed capacity until validation succeeds.
Rollback
If criteria fail, stop expansion:
-
Set
limits.cpu: 0on every NAPNodePoolto stop new NAP capacity. -
Add the
karpenter.azure.com/disable:NoScheduletaint to eachNodePool. -
Create or scale fixed pools to the verified capacity needed by the workloads.
-
Allow the scheduler and NAP disruption process to move workloads, honoring PDBs, until no NAP nodes remain. Do not delete nodes directly.
-
Disable NAP:
az aks update \ --name "<CLUSTER_NAME>" --resource-group "<RESOURCE_GROUP>" \ --node-provisioning-mode Manual