Exposing custom metrics pushed on Azure Monitor to Horizontal Pod Autoscaler

@Amarkumar Mishra - FT , Thank you for the question.

Please confirm if:

1. the Azure Kubernetes Metric Adapter was setup completely to expose your custom metric? For reference, please refer to this link (especially the point 3).
Setting up Azure Kubernetes Metric Adapte
Please check if this custom metric is getting exposed using the kubectl get --raw as mentioned in this example.

2. Once you are able to expose the custom metric though metric server, you may follow the steps available here for scaling up the deployment.

You may also refer to this link for more details. Please let us know if you have any questions.

@AnuragSingh-MSFT the Azure Metric adapter has been deployed & it it running in cluster.

In the documentation provided above, they are using an external metrics AvgRequestCountPerHealthyHost present inside Microsoft.Network. But in my case I wanted to use custom metrics present in Azure Monitor with following details

Metric Namespace: custom-connection-count  
Metric: total-connection-count  

My main query how should I expose this metrics to Azure metrics adapter. What must be the content of custom-metrics.yaml file. Plz refer below yaml code

apiVersion: azure.com/v1alpha2  
kind: <Should it be CustomMetric or ExternalMetric>  
metadata:  
  name: connection-count-metric  
spec:  
  type: azuremonitor  
  azure:  
    resourceGroup: My-resource-group  
    resourceName: My-resource-name  
    resourceProviderNamespace: <which resource proivder should be here for custom metrics>  
    resourceType: <which resource type should be here for custom metrics>  
  metric:  
    metricName: custom-connection-count/total-connection-count <is it correct>  
    aggregation: Average  

Let me know how should I prepare this yaml file so that I can expose my custom metrics to metrics adapter.

P.S.: The Azure documentation does not mention anything around exposing custom metrics (pushed from code to Azure monitor) to Azure metrics adapter, thus making it very difficult to implement such things.

@Amarkumar Mishra - FT , please give us some time as we are looking into it and will get back to you soon. Thanks