Muokkaa

Jaa

Secure credentials with linked services using the mssparkutils

  • Artikkeli

Accessing data from external sources is a common pattern. Unless the external data source allows anonymous access, chances are you need to secure your connection with a credential, secret, or connection string.

Azure Synapse Analytics uses Microsoft Entra passthrough by default for authentication between resources. If you need to connect to a resource using other credentials, use the mssparkutils directly. The mssparkutils package simplifies the process of retrieving SAS tokens, Microsoft Entra tokens, connection strings, and secrets stored in a linked service or from an Azure Key Vault.

Microsoft Entra passthrough uses permissions assigned to you as a user in Microsoft Entra ID, rather than permissions assigned to Synapse or a separate service principal. For example, if you want to use Microsoft Entra passthrough to access a blob in a storage account, then you should go to that storage account and assign blob contributor role to yourself.

When retrieving secrets from Azure Key Vault, we recommend creating a linked service to your Azure Key Vault. Ensure that the Synapse workspace managed service identity (MSI) has Secret Get privileges on your Azure Key Vault. Synapse will authenticate to Azure Key Vault using the Synapse workspace managed service identity. If you connect directly to Azure Key Vault without a linked service, authenticate using your user Microsoft Entra credential.

For more information, see linked services.

Usage

mssparkutils help for tokens and secrets

This function displays the help documentation for secrets and tokens management in Synapse.

mssparkutils.credentials.help()

mssparkutils.credentials.help()

Console.WriteLine(TokenLibrary.help());

Get result:

 getToken(audience: String, name: String): returns AAD token for a given audience, name (optional)
 isValidToken(token: String): returns true if token hasn't expired
 getConnectionStringOrCreds(linkedService: String): returns connection string or credentials for the linked service
 getFullConnectionString(linkedService: String): returns full connection string with credentials for the linked service
 getPropertiesAll(linkedService: String): returns all the properties of the linked service
 getSecret(akvName: String, secret: String, linkedService: String): returns AKV secret for a given AKV linked service, akvName, secret key using workspace MSI
 getSecret(akvName: String, secret: String): returns AKV secret for a given akvName, secret key using user credentials
 getSecretWithLS(linkedService: String, secret: String): returns AKV secret for a given linked service, secret key
 putSecret(akvName: String, secretName: String, secretValue: String): puts AKV secret for a given akvName, secretName
 putSecret(akvName: String, secretName: String, secretValue: String, linkedService: String): puts AKV secret for a given akvName, secretName
 putSecretWithLS(linkedService: String, secretName: String, secretValue: String): puts AKV secret for a given linked service, secretName

Access Azure Data Lake Storage Gen2

ADLS Gen2 Primary Storage

Accessing files from the primary Azure Data Lake Storage uses Microsoft Entra passthrough for authentication by default and doesn't require the explicit use of the mssparkutils. The identity used in the passthrough authentication differs based on a few factors. By default, interactive notebooks are executed using the user's identity, but it can be changed to the workspace managed service identity (MSI). Batch jobs and non-interactive executions of the notebook use the Workspace MSI.

val df = spark.read.csv("abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<FILE PATH>")
display(df.limit(10))

df = spark.read.csv('abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<FILE PATH>')
display(df.limit(10))

ADLS Gen2 storage with linked services

Azure Synapse Analytics provides an integrated linked services experience when connecting to Azure Data Lake Storage Gen2. Linked services can be configured to authenticate using an Account Key, Service Principal, Managed Identity, or Credential.

When the linked service authentication method is set to Account Key, the linked service authenticates using the provided storage account key, request a SAS key, and automatically apply it to the storage request using the LinkedServiceBasedSASProvider.

Synapse allows users to set the linked service for a particular storage account. This makes it possible to read/write data from multiple storage accounts in a single spark application/query. Once we set spark.storage.synapse.{source_full_storage_account_name}.linkedServiceName for each storage account that will be used, Synapse figures out which linked service to use for a particular read/write operation. However if our spark job only deals with a single storage account, we can omit the storage account name and use spark.storage.synapse.linkedServiceName.

Note

It is not possible to change the authentication method of the default ABFS storage container.

val sc = spark.sparkContext
val source_full_storage_account_name = "teststorage.dfs.core.windows.net"
spark.conf.set(s"spark.storage.synapse.$source_full_storage_account_name.linkedServiceName", "<LINKED SERVICE NAME>")
sc.hadoopConfiguration.set(s"fs.azure.account.auth.type.$source_full_storage_account_name", "SAS")
sc.hadoopConfiguration.set(s"fs.azure.sas.token.provider.type.$source_full_storage_account_name", "com.microsoft.azure.synapse.tokenlibrary.LinkedServiceBasedSASProvider")

val df = spark.read.csv("abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<FILE PATH>")

display(df.limit(10))

%%pyspark
# Set the required configs
source_full_storage_account_name = "teststorage.dfs.core.windows.net"
spark.conf.set(f"spark.storage.synapse.{source_full_storage_account_name}.linkedServiceName", "<lINKED SERVICE NAME>")
sc._jsc.hadoopConfiguration().set(f"fs.azure.account.auth.type.{source_full_storage_account_name}", "SAS")
sc._jsc.hadoopConfiguration().set(f"fs.azure.sas.token.provider.type.{source_full_storage_account_name}", "com.microsoft.azure.synapse.tokenlibrary.LinkedServiceBasedSASProvider")

# Python code
df = spark.read.csv('abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<DIRECTORY PATH>')

df.show()

When the linked service authentication method is set to Managed Identity or Service Principal, the linked service uses the Managed Identity or Service Principal token with the LinkedServiceBasedTokenProvider provider.

val sc = spark.sparkContext
val source_full_storage_account_name = "teststorage.dfs.core.windows.net"
spark.conf.set(s"spark.storage.synapse.$source_full_storage_account_name.linkedServiceName", "<LINKED SERVICE NAME>")
sc.hadoopConfiguration.set(s"fs.azure.account.oauth.provider.type.$source_full_storage_account_name", "com.microsoft.azure.synapse.tokenlibrary.LinkedServiceBasedTokenProvider") 
val df = spark.read.csv("abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<FILE PATH>")

display(df.limit(10))

%%pyspark
# Python code
source_full_storage_account_name = "teststorage.dfs.core.windows.net"
spark.conf.set(f"spark.storage.synapse.{source_full_storage_account_name}.linkedServiceName", "<LINKED SERVICE NAME>")
sc._jsc.hadoopConfiguration().set(f"fs.azure.account.oauth.provider.type.{source_full_storage_account_name}", "com.microsoft.azure.synapse.tokenlibrary.LinkedServiceBasedTokenProvider")

df = spark.read.csv('abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<DIRECTORY PATH>')

df.show()

Set authentication settings through spark configuration

Authentication settings can also be specified through spark configurations, instead of running spark statements. All spark configurations should be prefixed with spark. and all hadoop configurations should be prefixed with spark.hadoop..

Spark config name Config value
spark.storage.synapse.teststorage.dfs.core.windows.net.linkedServiceName LINKED SERVICE NAME
spark.hadoop.fs.azure.account.oauth.provider.type.teststorage.dfs.core.windows.net microsoft.azure.synapse.tokenlibrary.LinkedServiceBasedTokenProvider

ADLS Gen2 storage without linked services

Connect to ADLS Gen2 storage directly by using a SAS key. Use the ConfBasedSASProvider and provide the SAS key to the spark.storage.synapse.sas configuration setting. SAS tokens can be set at the container level, account level, or global. We do not recommend setting SAS keys at the global level, as the job will not be able to read/write from more than one storage account.

SAS configuration per storage container

%%spark
sc.hadoopConfiguration.set("fs.azure.account.auth.type.<ACCOUNT>.dfs.core.windows.net", "SAS")
sc.hadoopConfiguration.set("fs.azure.sas.token.provider.type", "com.microsoft.azure.synapse.tokenlibrary.ConfBasedSASProvider")
spark.conf.set("spark.storage.synapse.<CONTAINER>.<ACCOUNT>.dfs.core.windows.net.sas", "<SAS KEY>")

val df = spark.read.csv("abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<FILE PATH>")

display(df.limit(10))

%%pyspark

sc._jsc.hadoopConfiguration().set("fs.azure.account.auth.type.<ACCOUNT>.dfs.core.windows.net", "SAS")
sc._jsc.hadoopConfiguration().set("fs.azure.sas.token.provider.type", "com.microsoft.azure.synapse.tokenlibrary.ConfBasedSASProvider")
spark.conf.set("spark.storage.synapse.<CONTAINER>.<ACCOUNT>.dfs.core.windows.net.sas", "<SAS KEY>")

df = spark.read.csv('abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<FILE PATH>')

display(df.limit(10))

SAS configuration per storage account

%%spark
sc.hadoopConfiguration.set("fs.azure.account.auth.type.<ACCOUNT>.dfs.core.windows.net", "SAS")
sc.hadoopConfiguration.set("fs.azure.sas.token.provider.type", "com.microsoft.azure.synapse.tokenlibrary.ConfBasedSASProvider")
spark.conf.set("spark.storage.synapse.<ACCOUNT>.dfs.core.windows.net.sas", "<SAS KEY>")

val df = spark.read.csv("abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<FILE PATH>")

display(df.limit(10))

%%pyspark

sc._jsc.hadoopConfiguration().set("fs.azure.account.auth.type.<ACCOUNT>.dfs.core.windows.net", "SAS")
sc._jsc.hadoopConfiguration().set("fs.azure.sas.token.provider.type", "com.microsoft.azure.synapse.tokenlibrary.ConfBasedSASProvider")
spark.conf.set("spark.storage.synapse.<ACCOUNT>.dfs.core.windows.net.sas", "<SAS KEY>")

df = spark.read.csv('abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<FILE PATH>')

display(df.limit(10))

SAS configuration of all storage accounts

%%spark
sc.hadoopConfiguration.set("fs.azure.account.auth.type", "SAS")
sc.hadoopConfiguration.set("fs.azure.sas.token.provider.type", "com.microsoft.azure.synapse.tokenlibrary.ConfBasedSASProvider")
spark.conf.set("spark.storage.synapse.sas", "<SAS KEY>")

val df = spark.read.csv("abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<FILE PATH>")

display(df.limit(10))

%%pyspark

sc._jsc.hadoopConfiguration().set("fs.azure.account.auth.type", "SAS")
sc._jsc.hadoopConfiguration().set("fs.azure.sas.token.provider.type", "com.microsoft.azure.synapse.tokenlibrary.ConfBasedSASProvider")
spark.conf.set("spark.storage.synapse.sas", "<SAS KEY>")

df = spark.read.csv('abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<FILE PATH>')

display(df.limit(10))

Use MSAL to acquire tokens (using custom app credentials)

When the ABFS storage driver is configured to use MSAL directly for authentications, the provider doesn't cache tokens. This can result in reliability issues. We recommend using the ClientCredsTokenProvider is part of the Synapse Spark.

%%spark
val source_full_storage_account_name = "teststorage.dfs.core.windows.net"
sc.hadoopConfiguration.set("fs.azure.sas.token.provider.type", "com.microsoft.azure.synapse.tokenlibrary.ClientCredsTokenProvider")
spark.conf.set("fs.azure.account.oauth2.client.id.$source_full_storage_account_name", "<Entra AppId>")
spark.conf.set("fs.azure.account.oauth2.client.secret.$source_full_storage_account_name", "<Entra app secret>")
spark.conf.set("fs.azure.account.oauth2.client.endpoint.$source_full_storage_account_name", "https://login.microsoftonline.com/<tenantid>")

val df = spark.read.csv("abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<FILE PATH>")

display(df.limit(10))

%%pyspark
source_full_storage_account_name = "teststorage.dfs.core.windows.net"
sc._jsc.hadoopConfiguration().set("fs.azure.sas.token.provider.type", "com.microsoft.azure.synapse.tokenlibrary.ClientCredsTokenProvider")
spark.conf.set(f"fs.azure.account.oauth2.client.id.{source_full_storage_account_name}.linkedServiceName", "<Entra AppId>")
spark.conf.set(f"fs.azure.account.oauth2.client.secret.{source_full_storage_account_name}.linkedServiceName", "<Entra app secret>")
spark.conf.set(f"fs.azure.account.oauth2.client.endpoint.{source_full_storage_account_name}.linkedServiceName", "https://login.microsoftonline.com/<tenantid>")

df = spark.read.csv('abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<FILE PATH>')
display(df.limit(10))

ADLS Gen2 storage with SAS token (from Azure Key Vault)

Connect to ADLS Gen2 storage using a SAS token stored in Azure Key Vault secret.

%%spark
sc.hadoopConfiguration.set("fs.azure.account.auth.type", "SAS")
sc.hadoopConfiguration.set("fs.azure.sas.token.provider.type", "com.microsoft.azure.synapse.tokenlibrary.AkvBasedSASProvider")
spark.conf.set("spark.storage.synapse.akv", "<AZURE KEY VAULT NAME>")
spark.conf.set("spark.storage.akv.secret", "<SECRET KEY>")

val df = spark.read.csv("abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<FILE PATH>")

display(df.limit(10))

%%pyspark
sc._jsc.hadoopConfiguration().set("fs.azure.account.auth.type", "SAS")
sc._jsc.hadoopConfiguration().set("fs.azure.sas.token.provider.type", "com.microsoft.azure.synapse.tokenlibrary.AkvBasedSASProvider")
spark.conf.set("spark.storage.synapse.akv", "<AZURE KEY VAULT NAME>")
spark.conf.set("spark.storage.akv.secret", "<SECRET KEY>")

df = spark.read.csv('abfss://<CONTAINER>@<ACCOUNT>.dfs.core.windows.net/<FILE PATH>')

display(df.limit(10))

TokenLibrary for other linked services

To connect to other linked services, you can make a direct call to the TokenLibrary.

getConnectionString()

To retrieve the connection string, use the getConnectionString function and pass in the linked service name.

%%spark
// retrieve connectionstring from mssparkutils

mssparkutils.credentials.getFullConnectionString("<LINKED SERVICE NAME>")

%%pyspark
# retrieve connectionstring from mssparkutils

mssparkutils.credentials.getFullConnectionString("<LINKED SERVICE NAME>")

%%csharp
// retrieve connectionstring from TokenLibrary

using Microsoft.Spark.Extensions.Azure.Synapse.Analytics.Utils;

string connectionString = TokenLibrary.GetConnectionString(<LINKED SERVICE NAME>);
Console.WriteLine(connectionString);

getPropertiesAll()

The getPropertiesAll is a helper function available in Scala and Python to get all the properties of a linked service

%%pyspark
import json
# retrieve connectionstring from mssparkutils

json.loads(mssparkutils.credentials.getPropertiesAll("<LINKED SERVICE NAME>"))

The output will look like

{
    'AuthType': 'Key',
    'AuthKey': '[REDACTED]',
    'Id': None,
    'Type': 'AzureBlobStorage',
    'Endpoint': 'https://storageaccount.blob.core.windows.net/',
    'Database': None
}

GetSecret()

To retrieve a secret stored from Azure Key Vault, we recommend that you create a linked service to Azure Key Vault within the Synapse workspace. The Synapse workspace managed service identity will need to be granted GET Secrets permission to the Azure Key Vault. The linked service will use the managed service identity to connect to Azure Key Vault service to retrieve the secret. Otherwise, connecting directly to Azure Key Vault will use the user's Microsoft Entra credential. In this case, the user will need to be granted the Get Secret permissions in Azure Key Vault.

In government clouds, provide the fully qualified domain name of the keyvault.

mssparkutils.credentials.getSecret("<AZURE KEY VAULT NAME>", "<SECRET KEY>" [, <LINKED SERVICE NAME>])

To retrieve a secret from Azure Key Vault, use the mssparkutils.credentials.getSecret() function.


mssparkutils.credentials.getSecret("<AZURE KEY VAULT NAME>", "<SECRET KEY>", "<LINKED SERVICE NAME>")


mssparkutils.credentials.getSecret("<AZURE KEY VAULT NAME>", "<SECRET KEY>", "<LINKED SERVICE NAME>")

using Microsoft.Spark.Extensions.Azure.Synapse.Analytics.Utils;

string connectionString = TokenLibrary.GetSecret("<AZURE KEY VAULT NAME>", "<SECRET KEY>", "<LINKED SERVICE NAME>");
Console.WriteLine(connectionString);

Linked service connections supported from the Spark runtime

While Azure Synapse Analytics supports various linked service connections (from pipelines and other Azure products), not all of them are supported from the Spark runtime. Here is the list of supported linked services:

  • Azure Blob Storage
  • Azure AI services
  • Azure Cosmos DB
  • Azure Data Explorer
  • Azure Database for MySQL
  • Azure Database for PostgreSQL
  • Azure Data Lake Store (Gen1)
  • Azure Key Vault
  • Azure Machine Learning
  • Azure Purview
  • Azure SQL Database
  • Azure SQL Data Warehouse (Dedicated and Serverless)
  • Azure Storage

mssparkutils.credentials.getToken()

When you need an OAuth bearer token to access services directly, you can use the getToken method. The following resources are supported:

Service Name String literal to be used in API call
Azure Storage Storage
Azure Key Vault Vault
Azure Management AzureManagement
Azure SQL Data Warehouse (Dedicated and Serverless) DW
Azure Synapse Synapse
Azure Data Lake Store DataLakeStore
Azure Data Factory ADF
Azure Data Explorer AzureDataExplorer
Azure Database for MySQL AzureOSSDB
Azure Database for MariaDB AzureOSSDB
Azure Database for PostgreSQL AzureOSSDB

Unsupported linked service access from the Spark runtime

The following methods of accessing the linked services are not supported from the Spark runtime:

  • Passing arguments to parameterized linked service
  • Connections with user-assigned managed identities (UAMI)
  • Getting the bearer token to Keyvault resource when your Notebook / SparkJobDefinition runs as managed identity
    • As an alternative, instead of getting an access token, you can create a linked service to Keyvault and get the secret from your Notebook / batch job
  • For Azure Cosmos DB connections, key based access alone is supported. Token based access is not supported.

While running a notebook or a Spark job, requests to get a token / secret using a linked service might fail with an error message that indicates 'BadRequest'. This is often caused by a configuration issue with the linked service. If you see this error message, please check the configuration of your linked service. If you have any questions, contact Microsoft Azure Support at the Azure portal.

Related content