AML to AKS updates

aks-aml/README.md

@@ -1,26 +1,44 @@
 # AKS ALM Integration
 
-## Cluster Setup
+## Azure CLI and kubectl Setup
+
+```bash
+# This guides uses commands from two Azure CLI extensions: k8s-extension and ml
+
+az extension add --upgrade -n ml
+az extension add --upgrade -n k8s-extension
+
+# It also assumes you have kubectl installed. If not, you can run:
+
+az aks install-cli
+```
+
+## AKS Cluster and Azure ML Workspace Setup
 
 ```bash
 # Set environment variables
+
 RG=EphAKSAMLLab
 CLUSTER_NAME=amltarget
 LOC=westus3
-ACR_NAME=amllab${RANDOM}
 AML_WORKSPACE_NAME=amllab
+CLUSTER_AML_NAMESPACE=amltest
 
 # Create Resource Group
+
 az group create -g $RG -l $LOC
 
 # Create the cluster
-az aks create -g $RG -n $CLUSTER_NAME
+
+az aks create -g $RG -n $CLUSTER_NAME --node-count 5
 
 AKS_CLUSTER_ID=$(az aks show -g $RG -n $CLUSTER_NAME --query id -o tsv)
 
 az aks get-credentials -g $RG -n $CLUSTER_NAME
 
-kubectl create ns amltest
+kubectl create ns $CLUSTER_AML_NAMESPACE
+
+# Install the AML extension in the cluster
 
 az k8s-extension create \
 --name amlextension \
@@ -33,11 +51,13 @@ az k8s-extension create \
 
 az k8s-extension show \
 --name amlextension \
---cluster-type connectedClusters \
+--cluster-type managedClusters \
 --cluster-name $CLUSTER_NAME \
 --resource-group $RG
 
 
+# Create the AML workspace and attach to cluster
+
 az ml workspace create -n $AML_WORKSPACE_NAME -g $RG
 
 az ml compute attach \
@@ -47,36 +67,59 @@ az ml compute attach \
 --name k8s-compute \
 --resource-id "${AKS_CLUSTER_ID}" \
 --identity-type SystemAssigned \
---namespace amltest
+--namespace $CLUSTER_AML_NAMESPACE
 ```
 
 ## Deploy Inference Endpoint
 
 ```bash
-az extension add -n ml
-az extension update -n ml
+# Setup environment
 
 export ENDPOINT_NAME="testamlaksendpoint"
+export DEPLOYMENT_NAME="blue"
+export MODEL_NAME="sklearn-regression"
+
+# Prepare files to be used by the Azure ML Online Deployment.
+# Be sure all files created (endpoint.yaml, deployment.yaml) or
+# downloaded (sklearn_regression_model.pkl, score.py, conda.yaml)
+# are in the same directory. Run all `az ml online-endpoint create`
+# and `az ml online-deployment create` commands from that directory
+# as well since they have relative path dependencies.
+
+# Grant the identity running the CLI commands Storage Blob Data Contributor
+# on the Azure ML workspace's storage account.
+
+az role assignment create \
+--role "Storage Blob Data Contributor" \
+--assignee-object-id "$(az ad signed-in-user show --query id -o tsv)" \
+--assignee-principal-type "User" \
+--scope "$(az ml workspace show -n $AML_WORKSPACE_NAME -g $RG --query storage_account -o tsv)"
+
+# Create local directory for all needed files
 
 mkdir model
 cd model
 
-cat <<"EOF" > endpoint.yaml
-$schema: https://azuremlschemas.azureedge.net/latest/kubernetesOnlineEndpoint.schema.json
-name: my-endpoint
+# Create Azure ML Online Endpoint specification file
+
+cat <<EOF > endpoint.yaml
+\$schema: https://azuremlschemas.azureedge.net/latest/kubernetesOnlineEndpoint.schema.json
+name: ${ENDPOINT_NAME}
 compute: azureml:k8s-compute
 auth_mode: Key
 tags:
-  tag1: endpoint-tag1-value
+  modelName: ${MODEL_NAME}
 EOF
 
+# Create Azure ML Online Deployment specification file
+
 cat <<EOF >deployment.yaml
 \$schema: https://azuremlschemas.azureedge.net/latest/kubernetesOnlineDeployment.schema.json
-name: blue
+name: ${DEPLOYMENT_NAME}
 type: kubernetes
 endpoint_name: ${ENDPOINT_NAME}
 model:
-  path: ./
+  path: ./sklearn_regression_model.pkl
 code_configuration:
   code: ./
   scoring_script: score.py
@@ -94,26 +137,56 @@ resources:
     cpu: "0.2"
     memory: "200Mi"
 tags:
-  tag1: deployment-tag1-value
+  endpointName: ${ENDPOINT_NAME}
+  modelName: ${MODEL_NAME}
 instance_count: 1
 scale_settings:
   type: default
 EOF
 
+# Download required model files
+
+wget https://raw.githubusercontent.com/Azure/azureml-examples/refs/heads/main/sdk/python/endpoints/online/model-1/model/sklearn_regression_model.pkl
+
 wget https://raw.githubusercontent.com/Azure/azureml-examples/refs/heads/main/sdk/python/endpoints/online/model-1/onlinescoring/score.py
 
 wget https://raw.githubusercontent.com/Azure/azureml-examples/refs/heads/main/sdk/python/endpoints/online/model-1/environment/conda.yaml
 
+
+# Create the Azure ML Online Endpoint
+
 az ml online-endpoint create \
 -g $RG \
 -w $AML_WORKSPACE_NAME \
 -n $ENDPOINT_NAME \
 -f endpoint.yaml
 
+# Create the Azure ML Online Deployment and send all traffic to it
+
 az ml online-deployment create \
 -g $RG \
 -w $AML_WORKSPACE_NAME \
--n blue \
+-n $DEPLOYMENT_NAME \
 --endpoint $ENDPOINT_NAME \
+--all-traffic \
 -f deployment.yaml
+```
+
+## Test the Inference Endpoint
+
+```bash
+# Get the key required for the Authorization header
+
+SCORING_ACCESS_KEY=$(kubectl get onlineendpoint $ENDPOINT_NAME -n $CLUSTER_AML_NAMESPACE -o jsonpath={.spec.authKeys.primaryKey})
+
+# Get the endpoint URL for the request
+
+SCORING_ENDPOINT_URL=$(kubectl get onlineendpoint $ENDPOINT_NAME -n $CLUSTER_AML_NAMESPACE -o jsonpath={.status.scoringUri})
+
+# Post the test data to test the inference endpoint
+
+curl -X POST -H "Content-Type: application/json" \
+-H "Authorization: Bearer $SCORING_ACCESS_KEY" \
+--data "$(curl -s https://raw.githubusercontent.com/Azure/azureml-examples/refs/heads/main/sdk/python/endpoints/online/model-1/sample-request.json)" \
+--url $SCORING_ENDPOINT_URL
 ```