CLI (v2) Azure Arc-enabled Kubernetes online deployment YAML schema

APPLIES TO: Azure CLI ml extension v2 (current)

The source JSON schema can be found at https://azuremlschemas.azureedge.net/latest/kubernetesOnlineDeployment.schema.json.

Note

The YAML syntax detailed in this document is based on the JSON schema for the latest version of the ML CLI v2 extension. This syntax is guaranteed only to work with the latest version of the ML CLI v2 extension. You can find the schemas for older extension versions at https://azuremlschemasprod.azureedge.net/.

YAML syntax

Key Type Description Allowed values Default value
$schema string The YAML schema. If you use the Azure Machine Learning VS Code extension to author the YAML file, including $schema at the top of your file enables you to invoke schema and resource completions.
name string Required. Name of the deployment.

Naming rules are defined here.
description string Description of the deployment.
tags object Dictionary of tags for the deployment.
endpoint_name string Required. Name of the endpoint to create the deployment under.
model string or object The model to use for the deployment. This value can be either a reference to an existing versioned model in the workspace or an inline model specification.

To reference an existing model, use the azureml:<model-name>:<model-version> syntax.

To define a model inline, follow the Model schema.

As a best practice for production scenarios, you should create the model separately and reference it here.

This field is optional for custom container deployment scenarios.
model_mount_path string The path to mount the model in a custom container. Applicable only for custom container deployment scenarios. If the model field is specified, it's mounted on this path in the container.
code_configuration object Configuration for the scoring code logic.

This field is optional for custom container deployment scenarios.
code_configuration.code string Local path to the source code directory for scoring the model.
code_configuration.scoring_script string Relative path to the scoring file in the source code directory.
environment_variables object Dictionary of environment variable key-value pairs to set in the deployment container. You can access these environment variables from your scoring scripts.
environment string or object Required. The environment to use for the deployment. This value can be either a reference to an existing versioned environment in the workspace or an inline environment specification.

To reference an existing environment, use the azureml:<environment-name>:<environment-version> syntax.

To define an environment inline, follow the Environment schema.

As a best practice for production scenarios, you should create the environment separately and reference it here.
instance_type string The instance type used to place the inference workload. If omitted, the inference workload will be placed on the default instance type of the Kubernetes cluster specified in the endpoint's compute field. If specified, the inference workload will be placed on that selected instance type.

The set of instance types for a Kubernetes cluster is configured via the Kubernetes cluster custom resource definition (CRD), hence they aren't part of the Azure ML YAML schema for attaching Kubernetes compute.For more information, see Create and select Kubernetes instance types.
instance_count integer The number of instances to use for the deployment. Specify the value based on the workload you expect. This field is only required if you're using the default scale type (scale_settings.type: default).

instance_count can be updated after deployment creation using az ml online-deployment update command.
app_insights_enabled boolean Whether to enable integration with the Azure Application Insights instance associated with your workspace. false
scale_settings object The scale settings for the deployment. The two types of scale settings supported are the default scale type and the target_utilization scale type.

With the default scale type (scale_settings.type: default), you can manually scale the instance count up and down after deployment creation by updating the instance_count property.

To configure the target_utilization scale type (scale_settings.type: target_utilization), see TargetUtilizationScaleSettings for the set of configurable properties.
scale_settings.type string The scale type. default, target_utilization target_utilization
request_settings object Scoring request settings for the deployment. See RequestSettings for the set of configurable properties.
liveness_probe object Liveness probe settings for monitoring the health of the container regularly. See ProbeSettings for the set of configurable properties.
readiness_probe object Readiness probe settings for validating if the container is ready to serve traffic. See ProbeSettings for the set of configurable properties.
resources object Container resource requirements.
resources.requests object Resource requests for the container. See ContainerResourceRequests for the set of configurable properties.
resources.limits object Resource limits for the container. See ContainerResourceLimits for the set of configurable properties.

RequestSettings

Key Type Description Default value
request_timeout_ms integer The scoring timeout in milliseconds. 5000
max_concurrent_requests_per_instance integer The maximum number of concurrent requests per instance allowed for the deployment.

Do not change this setting from the default value unless instructed by Microsoft Technical Support or a member of the Azure ML team.
1
max_queue_wait_ms integer The maximum amount of time in milliseconds a request will stay in the queue. 500

ProbeSettings

Key Type Description Default value
period integer How often (in seconds) to perform the probe. 10
initial_delay integer The number of seconds after the container has started before the probe is initiated. Minimum value is 1. 10
timeout integer The number of seconds after which the probe times out. Minimum value is 1. 2
success_threshold integer The minimum consecutive successes for the probe to be considered successful after having failed. Minimum value is 1. 1
failure_threshold integer When a probe fails, the system will try failure_threshold times before giving up. Giving up in the case of a liveness probe means the container will be restarted. In the case of a readiness probe the container will be marked Unready. Minimum value is 1. 30

TargetUtilizationScaleSettings

Key Type Description Default value
type const The scale type target_utilization
min_instances integer The minimum number of instances to use. 1
max_instances integer The maximum number of instances to scale to. 1
target_utilization_percentage integer The target CPU usage for the autoscaler. 70
polling_interval integer How often the autoscaler should attempt to scale the deployment, in seconds. 1

ContainerResourceRequests

Key Type Description
cpu string The number of CPU cores requested for the container.
memory string The memory size requested for the container
nvidia.com/gpu string The number of Nvidia GPU cards requested for the container

ContainerResourceLimits

Key Type Description
cpu string The limit for the number of CPU cores for the container.
memory string The limit for the memory size for the container.
nvidia.com/gpu string The limit for the number of Nvidia GPU cards for the container

Remarks

The az ml online-deployment commands can be used for managing Azure Machine Learning Kubernetes online deployments.

Examples

Examples are available in the examples GitHub repository.

Next steps