Lake Formation unfiltered access for EMR Serverless - Amazon EMR
Using Lake Formation with full table access
Services or capabilities described in Amazon Web Services documentation might vary by Region. To see the differences applicable to the China Regions, see Getting Started with Amazon Web Services in China (PDF).

Lake Formation unfiltered access for EMR Serverless

With Amazon EMR releases 7.8.0 and higher, you can leverage Amazon Lake Formation with Glue Data Catalog where the job runtime role has full table permissions without the limitations of fine-grained access control. This capability allows you to read and write to tables that are protected by Lake Formation from your EMR Serverless Spark batch and interactive jobs. See the following sections to learn more about Lake Formation and how to use it with EMR Serverless.

Using Lake Formation with full table access

You can access Amazon Lake Formation protected Glue Data catalog tables from EMR Serverless Spark jobs or interactive sessions where the job's runtime role has full table access. You do not have to enable Amazon Lake Formation on the EMR Serverless application. When a Spark job is configured for Full Table Access (FTA), Amazon Lake Formation credentials are used to read/write S3 data for Amazon Lake Formation registered tables, while the job's runtime role credentials will be used to read/write tables not registered with Amazon Lake Formation.

Important

Do not enable Amazon Lake Formation for fine-grained access control. A job cannot simultaneously run Full Table Access (FTA) and Fine-Grained Access Control (FGAC) on the same EMR cluster or application.

Step 1: Enable Full Table Access in Lake Formation

To use Full Table Access (FTA) mode, you need to allow third-party query engines to access data without the IAM session tag validation in Amazon Lake Formation. To enable, follow the steps in Application integration for full table access.

Note

When accessing cross-account tables, full-table access must be enabled in both producer and consumer accounts. In the same manner, when accessing cross-region tables, this setting must be enabled in both producer and consumer regions.

Step 2: Setup IAM permissions for job runtime role

For read or write access to underlying data, in addition to Lake Formation permissions, a job runtime role needs the lakeformation:GetDataAccess IAM permission. With this permission, Lake Formation grants the request for temporary credentials to access the data.

The following is an example policy of how to provide IAM permissions to access a script in Amazon S3, uploading logs to S3, Amazon Glue API permissions, and permission to access Lake Formation.

JSON
{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Sid": "ScriptAccess",
      "Effect": "Allow",
      "Action": [
        "s3:GetObject"
      ],
      "Resource": [
        "arn:aws:s3:::*.amzn-s3-demo-bucket/scripts"
      ]
    },
    {
      "Sid": "LoggingAccess",
      "Effect": "Allow",
      "Action": [
        "s3:PutObject"
      ],
      "Resource": [
        "arn:aws:s3:::amzn-s3-demo-bucket/logs/*"
      ]
    },
    {
      "Sid": "GlueCatalogAccess",
      "Effect": "Allow",
      "Action": [
        "glue:Get*",
        "glue:Create*",
        "glue:Update*"
      ],
      "Resource": [
        "*"
      ]
    },
    {
      "Sid": "LakeFormationAccess",
      "Effect": "Allow",
      "Action": [
        "lakeformation:GetDataAccess"
      ],
      "Resource": [
        "*"
      ]
    }
  ]
}

Step 2.1 Configure Lake Formation permissions

  • Spark jobs that read data from S3 require Lake Formation SELECT permission.

  • Spark jobs that write/delete data in S3 require Lake Formation ALL (SUPER) permission.

  • Spark jobs that interact with Glue Data catalog require DESCRIBE, ALTER, DROP permission as appropriate.

For more information, see Granting permissions on Data Catalog resources.

Step 3: Initialize a Spark session for Full Table Access using Lake Formation

Prerequisites

Amazon Glue Data Catalog must be configured as a metastore to access Lake Formation tables.

Set the following settings to configure Glue catalog as a metastore:

--conf spark.sql.catalogImplementation=hive
--conf spark.hive.metastore.client.factory.class=com.amazonaws.glue.catalog.metastore.AWSGlueDataCatalogHiveClientFactory

For more information on enabling Data Catalog for EMR Serverless, see Metastore configuration for EMR Serverless.

To access tables registered with Amazon Lake Formation, the following configurations need to be set during Spark initialization to configure Spark to use Amazon Lake Formation credentials.

Hive
‐‐conf spark.hadoop.fs.s3.credentialsResolverClass=com.amazonaws.glue.accesscontrol.AWSLakeFormationCredentialResolver
--conf spark.hadoop.fs.s3.useDirectoryHeaderAsFolderObject=true 
--conf spark.hadoop.fs.s3.folderObject.autoAction.disabled=true
--conf spark.sql.catalog.skipLocationValidationOnCreateTable.enabled=true
--conf spark.sql.catalog.createDirectoryAfterTable.enabled=true
--conf spark.sql.catalog.dropDirectoryBeforeTable.enabled=true
Iceberg
--conf spark.sql.catalog.spark_catalog=org.apache.iceberg.spark.SparkSessionCatalog
--conf spark.sql.catalog.spark_catalog.warehouse=S3_DATA_LOCATION
--conf spark.sql.catalog.spark_catalog.client.region=REGION
--conf spark.sql.catalog.spark_catalog.type=glue
--conf spark.sql.catalog.spark_catalog.glue.account-id=ACCOUNT_ID
--conf spark.sql.catalog.spark_catalog.glue.lakeformation-enabled=true
--conf spark.sql.catalog.dropDirectoryBeforeTable.enabled=true

  • spark.hadoop.fs.s3.credentialsResolverClass=com.amazonaws.glue.accesscontrol.AWSLakeFormationCredentialResolver: Configure EMR Filesystem (EMRFS) to use Amazon Lake Formation S3 credentials for Lake Formation registered tables. If the table is not registered, use the job's runtime role credentials.

  • spark.hadoop.fs.s3.useDirectoryHeaderAsFolderObject=true and spark.hadoop.fs.s3.folderObject.autoAction.disabled=true: Configure EMRFS to use content type header application/x-directory instead of $folder$ suffix when creating S3 folders. This is required when reading Lake Formation tables, as Lake Formation credentials do not allow reading table folders with $folder$ suffix.

  • spark.sql.catalog.skipLocationValidationOnCreateTable.enabled=true: Configure Spark to skip validating the table location's emptiness before creation. This is necessary for Lake Formation registered tables, as Lake Formation credentials to verify the empty location are available only after Glue Data Catalog table creation. Without this configuration, the job's runtime role credentials will validate the empty table location.

  • spark.sql.catalog.createDirectoryAfterTable.enabled=true: Configure Spark to create the Amazon S3 folder after table creation in the Hive metastore. This is required for Lake Formation registered tables, as Lake Formation credentials to create the S3 folder are available only after Glue Data Catalog table creation.

  • spark.sql.catalog.dropDirectoryBeforeTable.enabled=true: Configure Spark to drop the S3 folder before table deletion in the Hive metastore. This is necessary for Lake Formation registered tables, as Lake Formation credentials to drop the S3 folder are not available after table deletion from the Glue Data Catalog.

  • spark.sql.catalog.<catalog>.glue.lakeformation-enabled=true: Configure Iceberg catalog to use Amazon Lake Formation S3 credentials for Lake Formation registered tables. If the table is not registered, use default environment credentials.

Configure full table access mode in SageMaker Unified Studio

To access Lake Formation registered tables from interactive Spark sessions in JupyterLab notebooks, you must use compatibility permission mode. Use the %%configure magic command to set up your Spark configuration. Choose the configuration based on your table type:

For Hive tables
%%configure -f
{
    "conf": {
        "spark.hadoop.fs.s3.credentialsResolverClass": "com.amazonaws.glue.accesscontrol.AWSLakeFormationCredentialResolver",
        "spark.hadoop.fs.s3.useDirectoryHeaderAsFolderObject": true,
        "spark.hadoop.fs.s3.folderObject.autoAction.disabled": true,
        "spark.sql.catalog.skipLocationValidationOnCreateTable.enabled": true,
        "spark.sql.catalog.createDirectoryAfterTable.enabled": true,
        "spark.sql.catalog.dropDirectoryBeforeTable.enabled": true
    }
}
For Iceberg tables
%%configure -f
{
    "conf": {
        "spark.sql.catalog.spark_catalog": "org.apache.iceberg.spark.SparkSessionCatalog",
        "spark.sql.catalog.spark_catalog.warehouse": "S3_DATA_LOCATION",
        "spark.sql.catalog.spark_catalog.client.region": "REGION",
        "spark.sql.catalog.spark_catalog.type": "glue",
        "spark.sql.catalog.spark_catalog.glue.account-id": "ACCOUNT_ID",
        "spark.sql.catalog.spark_catalog.glue.lakeformation-enabled": true,
        "spark.sql.catalog.dropDirectoryBeforeTable.enabled": true, 
    }
}

Replace the placeholders:

  • S3_DATA_LOCATION: Your S3 bucket path

  • REGION: Amazon region (e.g., us-east-1)

  • ACCOUNT_ID: Your Amazon account ID

Note

You must set these configurations before executing any Spark operations in your notebook.

Supported Operations

These operations will use Amazon Lake Formation credentials to access the table data.

  • CREATE TABLE

  • ALTER TABLE

  • INSERT INTO

  • INSERT OVERWRITE

  • UPDATE

  • MERGE INTO

  • DELETE FROM

  • ANALYZE TABLE

  • REPAIR TABLE

  • DROP TABLE

  • Spark datasource queries

  • Spark datasource writes

Note

Operations not listed above will continue to use IAM permissions to access table data.

Considerations

  • If a Hive table is created using a job that doesn’t have full table access enabled, and no records are inserted, subsequent reads or writes from a job with full table access will fail. This is because EMR Spark without full table access adds the $folder$ suffix to the table folder name. To resolve this, you can either:

    • Insert at least one row into the table from a job that does not have FTA enabled.

    • Configure the job that does not have FTA enabled to not use $folder$ suffix in folder name in S3. This can be achieved by setting Spark configuration spark.hadoop.fs.s3.useDirectoryHeaderAsFolderObject=true.

    • Create a S3 folder at the table location s3://path/to/table/table_name using the Amazon S3 console or Amazon S3 CLI.

  • Full Table Access works exclusively with EMR Filesystem (EMRFS). S3A filesystem is not compatible.

  • Full Table Access is supported for Hive and Iceberg tables. Support for Hudi and Delta tables has not yet been added.

  • Jobs referencing tables with Lake Formation Fine-Grained Access Control (FGAC) rules or Glue Data Catalog Views will fail. To query a table with an FGAC rules or a Glue Data Catalog View, you need to use the FGAC mode. You can enable FGAC mode by following the steps outlined in the Amazon documentation: Using EMR Serverless with Amazon Lake Formation for fine-grained access control.

  • Full table access does not support Spark Streaming.

  • When writing Spark DataFrame to a Lake Formation table, only APPEND mode is supported: df.write.mode("append").saveAsTable(table_name)