Simplify real-time analytics with zero-ETL from Amazon DynamoDB to Amazon SageMaker Lakehouse

At AWS re:Invent 2024, we launched a no code zero-ETL integration between Amazon DynamoDB and Amazon SageMaker Lakehouse, simplifying how organizations deal with knowledge analytics and AI workflows. This integration alleviates the normal challenges of constructing and sustaining advanced extract, remodel, and cargo (ETL) pipelines for reworking NoSQL knowledge into analytics-ready codecs, which beforehand required important time and sources whereas introducing potential system vulnerabilities. Organizations can now seamlessly mix the energy of DynamoDB in dealing with fast, concurrent transactions with rapid analytical processing via the zero-ETL integration. For instance, an ecommerce platform storing person session knowledge and cart info in DynamoDB can now analyze this knowledge in close to actual time with out constructing customized pipelines. Gaming firms utilizing DynamoDB for participant knowledge can immediately analyze person conduct as occasions happen, enabling real-time insights into sport stability and participant engagement patterns.

The zero-ETL functionality makes use of built-in change knowledge seize (CDC) to mechanically synchronize knowledge updates and schema modifications between DynamoDB and SageMaker Lakehouse tables. Through the use of Apache Iceberg format, the mixing offers dependable efficiency with ACID transaction assist and environment friendly large-scale knowledge dealing with. Information scientists can practice ML fashions on recent knowledge and knowledge analysts can generate experiences utilizing present info, with typical synchronization latency in minutes moderately than hours.

On this submit, we share the best way to arrange this zero-ETL integration from DynamoDB to your SageMaker Lakehouse surroundings.

Answer overview

We use a SageMaker Lakehouse catalog, AWS Lake Formation, Amazon Athena, AWS Glue, and Amazon SageMaker Unified Studio for this integration. The next is the reference knowledge stream diagram for the zero-ETL integration.

ref architecture

The workflow consists of the next elements:

The lately launched zero-ETL integration functionality inside the AWS Glue console permits direct integration between DynamoDB and SageMaker Lakehouse, storing knowledge in Iceberg format. This streamlined method opens up new prospects for knowledge groups by making a large-scale open and safe knowledge ecosystem with out conventional ETL processing overhead.
When constructing a SageMaker Lakehouse structure, you should use an Amazon Easy Storage Service (Amazon S3) primarily based managed catalog as your zero-ETL goal, offering seamless knowledge integration with out transformation overhead. This method creates a sturdy basis on your SageMaker Lakehouse implementation whereas sustaining the cost-effectiveness and scalability inherent to Amazon S3 storage, enabling environment friendly analytics and machine studying workflows.
Organizations can use a Redshift Managed Storage (RMS) primarily based managed catalog after they want high-performance SQL analytics and multi-table transactions. This method makes use of RMS for storage whereas sustaining knowledge within the Iceberg format, offering an optimum stability of efficiency and adaptability.
After you identify your Lakehouse infrastructure, you possibly can entry it via various analytics engines, together with AWS providers like Athena, Amazon Redshift, AWS Glue, and Amazon EMR as unbiased providers. For a extra streamlined expertise, SageMaker Unified Studio gives centralized analytics administration, the place you possibly can question your knowledge from a single unified interface.

Conditions

On this part, we stroll via the steps to arrange your resolution sources and make sure your permission settings.

Create a SageMaker Unified Studio area, challenge, and IAM function

Earlier than you start, you want an AWS Id and Entry Administration (IAM) function for enabling the zero-ETL integration. On this submit, we use SageMaker Unified Studio, which gives a unified knowledge platform expertise. It mechanically manages required Lake Formation permissions on knowledge and catalogs for you.

It’s a must to first create a SageMaker Unified Studio area, an administrative entity that controls person entry, permissions, and sources for groups working inside the SageMaker Unified Studio surroundings. Notice down the SageMaker Unified Studio URL after you create the area. You can be utilizing this URL later to log in to the SageMaker Unified Studio portal and question our knowledge throughout a number of engines.

Then, you create a SageMaker Unified Studio challenge, an built-in growth surroundings (IDE) that gives a unified expertise for knowledge processing, analytics, and AI growth. As a part of challenge creation, an IAM function is mechanically generated. This function will probably be used once you entry SageMaker Unified Studio later. For extra particulars on the best way to create a SageMaker Unified Studio challenge and area, discuss with An built-in expertise for all of your knowledge and AI with Amazon SageMaker Unified Studio.

Put together a pattern dataset inside DynamoDB

To implement this resolution, you want a DynamoDB desk that may both be used out of your current sources, or created utilizing the pattern knowledge file that you would be able to import from an S3 bucket. For this submit, we information you thru importing pattern knowledge from an S3 bucket into a brand new DynamoDB desk, offering a sensible basis for the ideas mentioned.

To create a pattern desk in DynamoDB, full the next steps:

Obtain the fictional ecommerce_customer_behavior.csv dataset. This dataset captures buyer conduct and interactions on an ecommerce platform.
On the Amazon S3 console, open the S3 bucket utilized by the SageMaker Unified Studio challenge.
Add the CSV file you downloaded.

Choose the uploaded file to view its particulars web page.

Copy the worth for S3 URI and make a remark of it; you’ll use this path for the following DynamoDB desk creation step.

Create a Dynamo DB desk

Full the next steps to create a DynamoDB desk from a file from Amazon S3, utilizing the import from Amazon S3 performance. Then you possibly can allow the settings on the DynamoDB desk required to allow zero-ETL integration.

On the DynamoDB console, choose Imports from S3 within the navigation pane.
Choose Import from S3.

Enter the S3 URI from earlier step for Supply S3 URL, choose CSV for Import file format, and choose Subsequent.

Present the desk identify as ecommerce_customer_behavior, the partition key as customer_id, and the type key as product_id, then choose Subsequent.

Use the default desk settings, then choose Subsequent to overview the small print.

Evaluation the settings and choose Import.

It is going to take a couple of minutes for the import standing to alter from Importing to Accomplished.

When the import is full, it’s best to be capable of see the desk created on the Tables web page.

Choose the ecommerce_customer_behavior desk and choose Edit PTIR.

Choose Activate cut-off date restoration and choose Save modifications.

That is required for organising zero-ETL utilizing DynamoDB as supply.
On the Backups tab, it’s best to see the standing for PITR as On.

Moreover, you want to use a desk coverage to allow entry for zero-ETL integration. On the Permissions tab, and duplicate the next code below Useful resource-based coverage for desk:

{
    "Model": "2012-10-17",
    "Assertion": [
        {
            "Sid": "TablePolicy01",
            "Effect": "Allow",
            "Principal": {
                "Service": "glue.amazonaws.com"
            },
            "Action": [
                "dynamodb:ExportTableToPointInTime",
                "dynamodb:DescribeExport",
                "dynamodb:DescribeTable"
            ],
            "Useful resource": "*"
        }
    ]
}

This coverage makes use of all of the sources, which shouldn’t be utilized in manufacturing workload. To deploy this setup in manufacturing, prohibit it to solely particular zero-ETL integration sources by including a situation to the resource-based coverage.

Now that you’ve used the Amazon S3 import methodology to load a CSV file to create a DynamoDB desk, you possibly can allow zero-ETL integration on the desk.

Validate permission settings

To validate if the catalog permission setting is suitable, full the next steps:

On the AWS Glue console, choose Databases within the navigation pane.

Verify for the database salesmarketing_XXX.

Choose Catalog settings within the navigation pane, and save the permissions.

The next code is an instance of permissions for catalog settings:

{
    "Model": "2012-10-17",
    "Assertion": [
        {
            "Effect": "Allow",
            "Principal": {
                "AWS": "arn:aws:iam:::root"
            },
            "Action": "glue:CreateInboundIntegration",
            "Resource": "arn:aws:glue:::database/salesmarketing_XXX"
        },
        {
            "Effect": "Allow",
            "Principal": {
                "Service": "glue.amazonaws.com"
            },
            "Action": "glue:AuthorizeInboundIntegration",
            "Resource": "arn:aws:glue:::database/salesmarketing_XXX"
        }
    ]
}

Now you’re able to create your zero-ETL integration.

Create a zero-ETL integration

Full the next steps to create a zero-ETL integration:

On the AWS Glue console, choose Zero-ETL integrations within the navigation pane.

Choose “Create zero-ETL integration” to create a brand new configuration.

Choose Amazon DynamoDB because the supply sort.

Underneath Supply particulars, choose ecommerce_customer_behavior for DynamoDB desk.

Underneath Goal particulars, present the next info:
1. For AWS account, choose Use the present account.
2. For Information warehouse or catalog, enter the account ID of your default catalog.
3. For Goal database, enter salesmarketing_XXX.
4. For Goal IAM function, enter datazone_usr_role_XXX.

Underneath Output settings, choose Unnest all fields and Use major keys from DynamoDB tables, depart Configure goal desk identify because the default worth (ecommerce_customer_behavior), then choose Subsequent.

Enter zetl-ecommerce-customer-behavior for Identify below Integration particulars, then choose Subsequent.

Choose Create and launch integration to launch the mixing.

The standing ought to be Creating after the mixing is efficiently initiated.
The standing will change to Lively in roughly a minute.

Confirm that the SageMaker Lakehouse desk exists. This course of would possibly take as much as quarter-hour to finish, as a result of the default refresh interval from DynamoDB is about to fifteen minutes.

Validate the SageMaker Lakehouse desk

Now you can question your SageMaker Lakehouse desk, created via zero-ETL integration, utilizing numerous question engines. Full the next steps to confirm you possibly can you see the desk in SageMaker Unified Studio:

Choose your challenge to view its particulars web page.

Choose Information within the navigation pane.

Confirm that you would be able to see the Iceberg desk within the SageMaker Lakehouse catalog.

Question with Athena

On this part, we present the best way to use Athena to question the SageMaker Lakehouse desk from SageMaker Unified Studio. On the challenge web page, find the ecommerce_customer_behavior desk within the catalog, and on the choices menu (three dots), choose Question with Athena.

This creates a SELECT question towards the SageMaker Lakehouse desk in a brand new window, and it’s best to see the question outcomes as proven within the following screenshot.

Question with Amazon Redshift

You can too question the SageMaker Lakehouse desk from SageMaker Unified Studio utilizing Amazon Redshift. Full the next steps:

Choose the connection on the highest proper.
Choose Redshift (Lakehouse) from the checklist of connections.
Choose the awsdatacatalog database.
Choose the salesmarketing schema.
Choose Select button.

The outcomes will probably be proven within the Amazon Redshift Question Editor.

Question with Amazon EMR Serverless

You’ll be able to question the Lakehouse desk utilizing Amazon EMR Serverless, which makes use of Apache Spark’s processing capabilities. Full the next steps:

On the challenge web page, choose Compute within the navigation pane.
Choose Add compute on the Information processing tab to create an EMR Serverless compute related to the challenge.

You’ll be able to create new compute sources or hook up with current sources. For this instance, choose Create new compute sources.

Choose EMR Serverless.

Enter a compute identify (for instance, Gross sales-Advertising), choose the newest launch of EMR Serverless, and choose Add compute.

It is going to take a while to create the compute.

It’s best to see the standing as Began for the compute. Now it’s prepared for use as your compute choice for querying via a Jupyter pocket book.

Choose the Construct menu and choose JupyterLab.

It is going to take a while to arrange the workspace for working JupyterLab.

After the Jupyter Lab house is about up, it’s best to see a web page much like the next screenshot.

Choose the brand new folder icon to create a brand new folder.

Identify the folder lakehouse_zetl_lab.

Navigate to the folder you simply created and create a pocket book below this folder.

Choose the pocket book Python3 (ipykernel) on the Launcher tab, and rename the pocket book to query_lakehouse_table.

You’ll be able to observe that the pocket book is displaying native Python as default language and compute. The 2 drop down menus present the connection sort and compute for the chosen connection sort, simply above the primary cell inside the Jupyter pocket book.

Choose PySpark because the connection, and choose the EMR Serverless software as compute.

Enter the next pattern code to question the desk utilizing Spark SQL:

import sys
from pyspark.sql import SparkSession
from pyspark.sql.capabilities import *

# Set the present database
spark.catalog.setCurrentDatabase("salesmarketing_XXX")

# Execute SQL question and retailer ends in DataFrame
df = spark.sql("choose * from ecommerce_customer_behavior restrict 10")

# Show the outcomes
df.present()

You’ll be able to see the Spark DataFrame outcomes.

Clear up

To keep away from incurring future costs, delete the SageMaker area, DynamoDB desk, AWS Glue sources, and different objects created from this submit.

Conclusion

This submit demonstrated how one can set up a zero-ETL connection from DynamoDB to SageMaker Lakehouse, making your knowledge obtainable in Iceberg format with out constructing customized knowledge pipelines. We confirmed how one can analyze this DynamoDB knowledge via numerous compute engines inside SageMaker Unified Studio. This streamlined method alleviates conventional knowledge motion complexities, and permits extra environment friendly knowledge evaluation workflows instantly out of your DynamoDB tables.

Check out this resolution on your personal use case, and share your suggestions within the feedback.

Concerning the authors

Narayani Ambashta is an Analytics Specialist Options Architect at AWS, specializing in the automotive and manufacturing sector, the place she guides strategic prospects in creating fashionable knowledge and AI methods. With over 15 years of cross-industry expertise, she focuses on huge knowledge structure, real-time analytics, and AI/ML applied sciences, serving to organizations implement fashionable knowledge architectures. Her experience spans throughout lakehouse, generative AI, and IoT platforms, enabling prospects to drive digital transformation initiatives. When not architecting fashionable options, she enjoys staying energetic via sports activities and yoga.

Raj Ramasubbu is a Senior Analytics Specialist Options Architect centered on huge knowledge and analytics and AI/ML with AWS. He helps prospects architect and construct extremely scalable, performant, and safe cloud-based options on AWS. Raj supplied technical experience and management in constructing knowledge engineering, huge knowledge analytics, enterprise intelligence, and knowledge science options for over 18 years previous to becoming a member of AWS. He helped prospects in numerous {industry} verticals like healthcare, medical units, life sciences, retail, asset administration, automobile insurance coverage, residential REIT, agriculture, title insurance coverage, provide chain, doc administration, and actual property.

Yadgiri Pottabhathini is a Senior Analytics Specialist Options Architect within the media and leisure sector. He focuses on helping enterprise prospects with their knowledge and analytics cloud transformation initiatives, whereas offering steerage on accelerating their Generative AI adoption via the event of information foundations and fashionable knowledge methods that leverage open-source frameworks and applied sciences.

Junpei Ozono is a Sr. Go-to-market (GTM) Information & AI options architect at AWS in Japan. He drives technical market creation for knowledge and AI options whereas collaborating with world groups to develop scalable GTM motions. He guides organizations in designing and implementing modern data-driven architectures powered by AWS providers, serving to prospects speed up their cloud transformation journey via fashionable knowledge and AI options. His experience spans throughout fashionable knowledge architectures together with Information Mesh, Information Lakehouse, and Generative AI, enabling prospects to construct scalable and modern options on AWS.