Link Azure Synapse Analytics and Azure Machine Learning workspaces and attach Apache Spark pools(deprecated)

Статия
08/28/2024

Warning

The Azure Synapse Analytics integration with Azure Machine Learning, available in Python SDK v1, is deprecated. Users can still use Synapse workspace, registered with Azure Machine Learning, as a linked service. However, a new Synapse workspace can no longer be registered with Azure Machine Learning as a linked service. We recommend use of serverless Spark compute and attached Synapse Spark pools, available in CLI v2 and Python SDK v2. For more information, visit https://aka.ms/aml-spark.

In this article, you learn how to create a linked service that links your Azure Synapse Analytics workspace and Azure Machine Learning workspace.

With an Azure Machine Learning workspace, linked with an Azure Synapse workspace, you can attach an Apache Spark pool, powered by Azure Synapse Analytics, as a dedicated compute resource. You can use this resource for data wrangling at scale, or you can conduct model training - all from the same Python notebook.

You can link your ML workspace and Synapse workspace with the Python SDK or the Azure Machine Learning studio. You can also link workspaces, and attach a Synapse Spark pool, with a single Azure Resource Manager (ARM) template.

Prerequisites

Link workspaces with the Python SDK

Important

To successfully link to the Synapse workspace, you must be granted the Owner role of the Synapse workspace. Check your access in the Azure portal.

If you are only a Contributor to the Synapse workspace, and you don't have an Owner for that Synapse workspace, you can only use existing linked services. For more information, visit Retrieve and use an existing linked service.

This code employs the LinkedService and SynapseWorkspaceLinkedServiceConfiguration classes, to

Link your machine learning workspace ws with your Azure Synapse workspace
Register your Synapse workspace with Azure Machine Learning as a linked service

import datetime  
from azureml.core import Workspace, LinkedService, SynapseWorkspaceLinkedServiceConfiguration

# Azure Machine Learning workspace
ws = Workspace.from_config()

#link configuration 
synapse_link_config = SynapseWorkspaceLinkedServiceConfiguration(
    subscription_id=ws.subscription_id,
    resource_group= 'your resource group',
    name='mySynapseWorkspaceName')

# Link workspaces and register Synapse workspace in Azure Machine Learning
linked_service = LinkedService.register(workspace = ws,              
                                            name = 'synapselink1',    
                                            linked_service_config = synapse_link_config)

Important

Managed identity system_assigned_identity_principal_id is created for each linked service. You must grant this managed identity the Synapse Apache Spark Administrator role of the Synapse workspace before you start your Synapse session. For more information, visit How to manage Azure Synapse RBAC assignments in Synapse Studio.

To find the system_assigned_identity_principal_id of a specific linked service, use LinkedService.get('<your-mlworkspace-name>', '<linked-service-name>').

Manage linked services

View all the linked services associated with your machine learning workspace:

LinkedService.list(ws)

To unlink your workspaces, use the unregister() method:

linked_service.unregister()

Link workspaces via studio

Link your machine learning workspace and Synapse workspace via the Azure Machine Learning studio:

Sign in to the Azure Machine Learning studio
Select Linked Services in the Manage section of the left pane
Select Add integration

On the Link workspace form, populate the fields

Field	Description
Name	Provide a name for your linked service. References to this specific linked service use this name
Subscription name	Select the name of your subscription associated with your machine learning workspace
Synapse workspace	Select the Synapse workspace to which you want to link

Select Next to open the Select Spark pools (optional) form. On this form, you select which Synapse Spark pool to attach to your workspace
Select Next to open the Review form, and check your selections
Select Create to complete the linked service creation process

Get an existing linked service

Before you can attach a dedicated compute for data wrangling, you must have a machine learning workspace linked to an Azure Synapse Analytics workspace. We refer to this workspace as a linked service. Retrieval and use of an existing linked service requires User or Contributor permissions to the Azure Synapse Analytics workspace.

This example retrieves an existing linked service - synapselink1 - from the workspace ws, with the get() method:

from azureml.core import LinkedService

linked_service = LinkedService.get(ws, 'synapselink1')

Attach Synapse Spark pool as a compute

Once you retrieve the linked service, attach a Synapse Apache Spark pool as a dedicated compute resource for your data wrangling tasks. You can attach Apache Spark pools with

Azure Machine Learning studio
Azure Resource Manager (ARM) templates
The Azure Machine Learning Python SDK

Attach a pool via the studio

Sign in to the Azure Machine Learning studio
Select Linked Services in the Manage section of the left pane
Select your Synapse workspace
Select Attached Spark pools on the top left
Select Attach
Select your Apache Spark pool from the list and provide a name
1. This list identifies the available Synapse Spark pools that can be attached to your compute
2. To create a new Synapse Spark pool, see Quickstart: Create a new serverless Apache Spark pool using the Azure portal
Select Attach selected

Attach a pool with the Python SDK

You can also employ the Python SDK to attach an Apache Spark pool, as shown in this code example:

from azureml.core.compute import SynapseCompute, ComputeTarget

attach_config = SynapseCompute.attach_configuration(linked_service, #Linked synapse workspace alias
                                                    type='SynapseSpark', #Type of assets to attach
                                                    pool_name=synapse_spark_pool_name) #Name of Synapse spark pool 

synapse_compute = ComputeTarget.attach(workspace= ws,                
                                       name= synapse_compute_name, 
                                       attach_configuration= attach_config
                                      )

synapse_compute.wait_for_completion()

Verify the Apache Spark pool is attached.

ws.compute_targets['Synapse Spark pool alias']

This code

Configures the SynapseCompute with
1. The LinkedService, linked_service that you either created or retrieved in the previous step
2. The type of compute target you want to attach - in this case, SynapseSpark
3. The name of the Apache Spark pool. The name must match an existing Apache Spark pool that exists in your Azure Synapse Analytics workspace
Creates a machine learning ComputeTarget by passing in
1. The machine learning workspace you want to use, ws
2. The name you'd like to use to refer to the compute within the Azure Machine Learning workspace
3. The attach_configuration you specified when you configured your Synapse Compute
  1. The call to ComputeTarget.attach() is asynchronous, so the sample execution is blocked until the call completes

Споделяне чрез