AKS platform metric alert creation fails for node_memory_working_set_percentage although metric definition exists

We can no longer create Azure Monitor metric alerts for the AKS platform metric node_memory_working_set_percentage.

Metric namespace:

Microsoft.ContainerService/managedClusters

Metric name:

node_memory_working_set_percentage

Affected resource type:

Microsoft.ContainerService/managedClusters

The metric definition exists for this AKS resource. This command returns the metric successfully:

az monitor metrics list-definitions --resource "$AKS_ID" --namespace Microsoft.ContainerService/managedClusters --query "[?name.value=='node_memory_working_set_percentage']" -o jsonc

It shows:

• category: Nodes PREVIEW
• dimensions: node and nodepool
• supported aggregations: Maximum and Average
• primaryAggregationType: Average
• unit: Percent

But creating the metric alert fails:

az monitor metrics alert create \

-g "$ALERT_RG" \

-n "ha-cloud-aks_node_memory_working_set-dev620" \

--scopes "$AKS_ID" \

--severity 3 \

--evaluation-frequency 1m \

--window-size 5m \

--condition "avg Microsoft.ContainerService/managedClusters.node_memory_working_set_percentage greater than 100 with skipmetricvalidation" \

--action "$ACTION_GROUP_ID"

Actual CLI condition used greater-than operator. The portal form rejects angle brackets, so I spell it out here.

Error:

BadRequest: The metric with the name node_memory_working_set_percentage does not exist.

Activity IDs:

4ec17149-f5c3-4132-84fb-40fb8c9889bd

1473fee4-e26f-44c6-a54b-ed50f3ba49da

The same failure occurs from Terraform azurerm_monitor_metric_alert, even with skip_metric_validation set to true.

Things tried:

• Average aggregation
• Maximum aggregation
• with skipmetricvalidation
• no dimension filter
• where node includes wildcard
• where nodepool includes wildcard
• explicit target resource type and region
• same AKS resource with node_memory_rss_percentage

Control test:

A similar alert using node_memory_rss_percentage can be created successfully on the same AKS resource.

Expected behavior:

Because the metric definition exists and the metric is documented for AKS, metric alert creation should accept node_memory_working_set_percentage, or the metric should not appear in list-definitions.

This looks like an Azure Monitor metric alert backend regression for the AKS preview node metric node_memory_working_set_percentage.