Amazon DynamoDB helps you capture high-velocity data such as clickstream data to form customized user profiles and online order transaction data to develop customer order fulfillment applications, improve customer satisfaction, and get insights into sales revenue to create a promotional offer for the customer. It’s essential to store these data points in a centralized data lake, which can be transformed, analyzed, and combined with diverse organizational datasets to derive meaningful insights and make predictions.

A popular use case in order management is receiving, tracking, and fulfilling customer orders. The order management process begins when an order is placed and ends when the customer receives their package. When storing high-velocity order transaction data in DynamoDB, you can use Amazon Kinesis streaming to extract data and store it in a centralized data lake built on Amazon Simple Storage Service (Amazon S3).

Amazon Kinesis Data Streams for DynamoDB helps you to publish item-level changes in any DynamoDB table to a Kinesis data stream of your choice. Additionally, you can take advantage of this feature for use cases that require longer data retention on the stream and fan out to multiple concurrent stream readers. You also can integrate with Amazon Kinesis Data Analytics or Amazon Kinesis Data Firehose to publish data to downstream destinations such as Amazon Elasticsearch Service (Amazon ES), Amazon Redshift, or Amazon S3.

In this post, you use Kinesis Data Streams for DynamoDB and take advantage of managed streaming delivery of DynamoDB data to other Kinesis Data Stream by simply enabling Kinesis streaming connection from Amazon DynamoDB console. To process DynamoDB events from Kinesis, you have multiple options: Amazon Kinesis Client Library (KCL) applications, Lambda, Kinesis Data Analytics for Apache Flink, and Kinesis Data Firehose. In this post, you use Kinesis Data Firehose to save the raw data in the S3 data lake and Apache Hudi to batch process the data.

Architecture

The following diagram illustrates the order processing system architecture.

bdb1053 dta lake kds ddb hudi 1

In this architecture, users buy products in online retail shops and internally create an order transaction stored in DynamoDB. The order transaction data is ingested to the data lake and stored in the raw data layer. To achieve this, you enable Kinesis Data Streams for DynamoDB and use Kinesis Data Firehose to store data in Amazon S3. You use Lambda to transform the data from the delivery stream to remove unwanted data and finally store it in Parquet format. Next, you batch process the raw data and store it back in the Hudi dataset in the S3 data lake. You can then use Amazon Athena to do sales analysis. You build this entire data pipeline in a serverless manner.

Prerequisites

Complete the following steps to create AWS resources to build a data pipeline as mentioned in the architecture. For this post, we use the AWS Region us-west-1.

  1. On the Amazon Elastic Compute Cloud (Amazon EC2) console, create a keypair.
  2. Download the data files, Amazon EMR cluster, and Athena DDL code from GitHub.
  3. Deploy the necessary Amazon resources using the provided AWS CloudFormation template.
    LaunchStack
  4. For Stack name, enter a stack name of your choice.
  5. For Keypair name, choose a key pair.

A key pair is required to connect to the EMR cluster nodes. For more information, see Use an Amazon EC2 Key Pair for SSH Credentials.
bdb1053 dta lake kds ddb hudi 2

  1. Keep the remaining default parameters.
  2. Acknowledge that AWS CloudFormation might create AWS Identity and Access Management (IAM) resources.

For more information about IAM, see Resources to learn more about IAM.

  1. Choose Create stack.

bdb1053 dta lake kds ddb hudi 3

You can check the Resources tab for the stack after the stack is created.

bdb1053 dta lake kds ddb hudi 4

The following table summarizes the resources that you created, which you use to build the data pipeline and analysis.

Logical IDPhysical IDType
DeliveryPolicykines-Deli-*AWS::IAM::Policy
DeliveryRolekinesis-hudi-DeliveryRole-*AWS::IAM::Role
Deliverystreamkinesis-hudi-Deliverystream-*AWS::KinesisFirehose::DeliveryStream
DynamoDBTableorder_transaction_*AWS::DynamoDB::Table
EMRClusterServiceRolekinesis-hudi-EMRClusterServiceRole-*AWS::IAM::Role
EmrInstanceProfilekinesis-hudi-EmrInstanceProfile-*AWS::IAM::InstanceProfile
EmrInstanceRolekinesis-hudi-EmrInstanceRole-*AWS::IAM::Role
GlueDatabasegluedatabase-*AWS::Glue::Database
GlueTablegluetable-*AWS::Glue::Table
InputKinesisStreamorder-data-stream-*AWS::Kinesis::Stream
InternetGatewayigw-*AWS::EC2::InternetGateway
InternetGatewayAttachmentkines-Inter-*AWS::EC2::VPCGatewayAttachment
MyEmrClusterAWS::EMR::Cluster
ProcessLambdaExecutionRolekinesis-hudi-ProcessLambdaExecutionRole-*AWS::IAM::Role
ProcessLambdaFunctionkinesis-hudi-ProcessLambdaFunction-*AWS::Lambda::Function
ProcessedS3Bucketkinesis-hudi-processeds3bucket-*AWS::S3::Bucket
PublicRouteTableAWS::EC2::RouteTable
PublicSubnet1AWS::EC2::Subnet
PublicSubnet1RouteTableAssociationAWS::EC2::SubnetRouteTableAssociation
PublicSubnet2AWS::EC2::Subnet
PublicSubnet2RouteTableAssociationAWS::EC2::SubnetRouteTableAssociation
RawS3Bucketkinesis-hudi-raws3bucket-*AWS::S3::Bucket
S3Bucketkinesis-hudi-s3bucket-*AWS::S3::Bucket
SourceS3Bucketkinesis-hudi-sources3bucket-*AWS::S3::Bucket
VPCvpc-*AWS::EC2::VPC

Enable Kinesis streaming for DynamoDB

AWS recently launched Kinesis Data Streams for DynamoDB so you can send data from DynamoDB to Kinesis data streams. You can use the AWS Command Line Interface (AWS CLI) or the AWS Management Console to enable this feature.

To enable this feature from the console, complete the following steps:

  1. On the DynamoDB console, choose the table you created in the CloudFormation stack earlier (it begins with the prefix order_transaction_).
  2. On the Overview tab, choose Manage streaming to Kinesis.
    bdb1053 dta lake kds ddb hudi 5
  3. Choose your input stream (it starts with order-data-stream-).
    bdb1053 dta lake kds ddb hudi 6
  4. Choose Enable.
    bdb1053 dta lake kds ddb hudi 7
  5. Choose Close.
  6. Make sure that stream enabled is set to Yes.

Populate the sales order transaction dataset

To replicate a real-life use case, you need an online retail application. For this post, you upload raw data files in the S3 bucket and use a Lambda function to upload the data in DynamoDB. You can download the order data CSV files from the AWS Sample GitHub repository. Complete the following steps to upload the data in DynamoDB:

  1. On the Amazon S3 console, choose the bucket <stack-name>-sourcess3bucket-*.
  2. Choose Upload.
    bdb1053 dta lake kds ddb hudi 8
  3. Choose Add files.
  4. Choose the order_data_09_02_2020.csv and order_data_10_02_2020.csv files.
  5. Choose Upload.
  6. On the Lambda console, choose the function <stack-name>-CsvToDDBLambdaFunction-*.
    bdb1053 dta lake kds ddb hudi 9
  7. Choose Test.
    bdb1053 dta lake kds ddb hudi 10
  8. For Event template, enter an event name.
  9. Choose Create.
    bdb1053 dta lake kds ddb hudi 11
  10. Choose Test.

This runs the Lambda function and loads the CSV file order_data_09_02_2020.csv to the DynamoDB table.

  1. Wait until the message appears that the function ran successfully.

bdb1053 dta lake kds ddb hudi 12

You can now view the data on the DynamoDB console, in the details page for your table.

bdb1053 dta lake kds ddb hudi 13

Because you enabled the Kinesis data stream in the DynamoDB table, it starts streaming the data to Amazon S3. You can check the data by viewing the bucket on the Amazon S3 console. The following screenshot shows that a Parquet file is under the prefix in the bucket.

bdb1053 dta lake kds ddb hudi 14

Use Apache Hudi with Amazon EMR

Now it’s time to process the streaming data using Hudi.

  1. Log in to the Amazon EMR leader node.

You can use the key pair you chose in the security options to SSH into the leader node.

  1. Use the following bash command to start the Spark shell to use it with Apache Hudi:
spark-shell --conf "spark.serializer=org.apache.spark.serializer.KryoSerializer" --conf "spark.sql.hive.convertMetastoreParquet=false" --jars /usr/lib/hudi/hudi-spark-bundle.jar,/usr/lib/spark/external/lib/spark-avro.jar

The Amazon EMR instance looks like the following screenshot.

bdb1053 dta lake kds ddb hudi 15

  1. You can use the following Scala code to import the order transaction data from the S3 data lake to a Hudi dataset using the copy-on-write storage type. Change inputDataPath as per file path in <stack-name>-raws3bucket-* in your environment, and replace the bucket name in hudiTablePath as <stack-name>- processeds3bucket-*.
import org.apache.spark.sql.SaveMode
import org.apache.spark.sql.functions._
import org.apache.hudi.DataSourceWriteOptions
import org.apache.hudi.config.HoodieWriteConfig
import org.apache.hudi.hive.MultiPartKeysValueExtractor //Set up various input values as variables
val inputDataPath = "s3://kinesis-hudi-raws3bucket-1p6nszvqd9awz/2021/02/01/15/"
val hudiTableName = "order_hudi_cow"
val hudiTablePath = "s3://kinesis-hudi-processeds3bucket-yfc6843vmq1o/" + hudiTableName // Set up our Hudi Data Source Options
val hudiOptions = Map[String,String]( DataSourceWriteOptions.RECORDKEY_FIELD_OPT_KEY -> "order_id", DataSourceWriteOptions.RECORDKEY_FIELD_OPT_KEY -> "time_stamp", DataSourceWriteOptions.PARTITIONPATH_FIELD_OPT_KEY -> "transaction_date", HoodieWriteConfig.TABLE_NAME -> hudiTableName, DataSourceWriteOptions.OPERATION_OPT_KEY -> DataSourceWriteOptions.INSERT_OPERATION_OPT_VAL, DataSourceWriteOptions.PRECOMBINE_FIELD_OPT_KEY -> "time_stamp", DataSourceWriteOptions.HIVE_SYNC_ENABLED_OPT_KEY -> "true", DataSourceWriteOptions.HIVE_TABLE_OPT_KEY -> hudiTableName, DataSourceWriteOptions.HIVE_PARTITION_FIELDS_OPT_KEY -> "transaction_date", DataSourceWriteOptions.HIVE_PARTITION_EXTRACTOR_CLASS_OPT_KEY -> classOf[MultiPartKeysValueExtractor].getName) // Read data from S3 and create a DataFrame with Partition and Record Key
val inputDF = spark.read.format("parquet").load(inputDataPath) // Write data into the Hudi dataset
inputDF.write.format("org.apache.hudi").options(hudiOptions).mode(SaveMode.Append).save(hudiTablePath)

For more information about DataSourceWriteOptions, see Work with a Hudi Dataset.

  1. In the Spark shell, you can now count the total number of records in the Apache Hudi dataset:
scala> inputDF.count()
res1: Long = 1000

bdb1053 dta lake kds ddb hudi 16

You can check the processed Apache Hudi dataset in the S3 data lake via the Amazon S3 console. The following screenshot shows the prefix order_hudi_cow is in <stack-name>- processeds3bucket-*.

bdb1053 dta lake kds ddb hudi 17

When navigating into the order_hudi_cow prefix, you can find a list of Hudi datasets that are partitioned using the transaction_date key—one for each date in our dataset.

bdb1053 dta lake kds ddb hudi 18

Let’s analyze the data stored in Amazon S3 using Athena.

Analyze the data with Athena

To analyze your data, complete the following steps:

  1. On the Athena console, create the database order_db using the following command:
create database order_db;

You use this database to create all the Athena tables.

  1. Create your table using the following command (replace the S3 bucket name with <stack-name>- processeds3bucket* created in your environment):
    CREATE EXTERNAL TABLE order_transaction_cow ( `_hoodie_commit_time` string, `_hoodie_commit_seqno` string, `_hoodie_record_key` string, `_hoodie_partition_path` string, `_hoodie_file_name` string, `order_id` string, `item_id` string, `customer_id` string, `product` string, `amount` decimal(3,1), `currency` string, `time_stamp` string ) PARTITIONED BY ( `transaction_date` string)
    ROW FORMAT SERDE 'org.apache.hadoop.hive.ql.io.parquet.serde.ParquetHiveSerDe'
    STORED AS INPUTFORMAT 'org.apache.hudi.hadoop.HoodieParquetInputFormat'
    OUTPUTFORMAT 'org.apache.hadoop.hive.ql.io.parquet.MapredParquetOutputFormat'
    LOCATION 's3://kinesis-hudi-processeds3bucket-yfc6843vmq1o/order_hudi_cow'

    bdb1053 dta lake kds ddb hudi 19

  2. Add partitions by running the following query on the Athena console:
    ALTER TABLE order_transaction_cow ADD
    PARTITION (transaction_date = '2020-09-02') LOCATION 's3://kinesis-hudi-processeds3bucket-yfc6843vmq1o/order_hudi_cow/2020-09-02/';

  3. Check the total number of records in the Hudi dataset with the following query:
    SELECT count(*) FROM "order_db"."order_transaction_cow";

It should return a single row with a count of 1,000.

Now check the record that you want to update.

 

  1. Run the following query on the Athena console:
SELECT * FROM "order_db"."order_transaction_cow"
where order_id ='3801'
and item_id ='23'
and transaction_date ='2020-09-02';

The output should look like the following screenshot. Note down the value of product and amount.

bdb1053 dta lake kds ddb hudi 20

Analyze the change data capture

Now let’s test the change data capture (CDC) in streaming. Let’s take an example where the customer changed an existing order. We load the order_data_10_02_2020.csv file, where order_id 3801 has a different product and amount.

bdb1053 dta lake kds ddb hudi 21

To test the CDC feature, complete the following steps:

  1. On the Lambda console, choose the stack <stack-name>-CsvToDDBLambdaFunction-*.
  2. In the Environment variables section, choose Edit.
    bdb1053 dta lake kds ddb hudi 22
  3. For key, enter order_data_10_02_2020.csv.
  4. Choose Save.

bdb1053 dta lake kds ddb hudi 23

You can see another prefix has been created in <stack-name>-raws3bucket-*.

bdb1053 dta lake kds ddb hudi 24

  1. In Amazon EMR, run the following code in the Scala shell prompt to update the data (change inputDataPath to the file path in <stack-name>-raws3bucket-* and hudiTablePath to <stack-name>- processeds3bucket-*):
    import org.apache.spark.sql.SaveMode
    import org.apache.spark.sql.functions._
    import org.apache.hudi.DataSourceWriteOptions
    import org.apache.hudi.config.HoodieWriteConfig
    import org.apache.hudi.hive.MultiPartKeysValueExtractor //Set up various input values as variables
    val inputDataPath = "s3://kinesis-hudi-raws3bucket-1p6nszvqd9awz/2021/02/01/18/"
    val hudiTableName = "order_hudi_cow"
    val hudiTablePath = "s3://kinesis-hudi-processeds3bucket-yfc6843vmq1o/" + hudiTableName // Set up our Hudi Data Source Options
    val hudiOptions = Map[String,String]( DataSourceWriteOptions.RECORDKEY_FIELD_OPT_KEY -> "order_id", DataSourceWriteOptions.RECORDKEY_FIELD_OPT_KEY -> "time_stamp", DataSourceWriteOptions.PARTITIONPATH_FIELD_OPT_KEY -> "transaction_date", HoodieWriteConfig.TABLE_NAME -> hudiTableName, DataSourceWriteOptions.OPERATION_OPT_KEY -> DataSourceWriteOptions.INSERT_OPERATION_OPT_VAL, DataSourceWriteOptions.PRECOMBINE_FIELD_OPT_KEY -> "time_stamp", DataSourceWriteOptions.HIVE_SYNC_ENABLED_OPT_KEY -> "true", DataSourceWriteOptions.HIVE_TABLE_OPT_KEY -> hudiTableName, DataSourceWriteOptions.HIVE_PARTITION_FIELDS_OPT_KEY -> "transaction_date", DataSourceWriteOptions.HIVE_PARTITION_EXTRACTOR_CLASS_OPT_KEY -> classOf[MultiPartKeysValueExtractor].getName) // Read data from S3 and create a DataFrame with Partition and Record Key
    val inputDF = spark.read.format("parquet").load(inputDataPath) // Write data into the Hudi dataset
    inputDF.write.format("org.apache.hudi").options(hudiOptions).mode(SaveMode.Append).save(hudiTablePath
    

  2. Run the following query on the Athena console to check for the change to the total number of records as 1,000:
    SELECT count(*) FROM "order_db"."order_transaction_cow";

  3. Run the following query on the Athena console to test for the update:
SELECT * FROM "order_db"."order_transaction_cow"
where order_id ='3801'
and item_id ='23'
and transaction_date ='2020-09-02';

The following screenshot shows that the product and amount values for the same order are updated.

bdb1053 dta lake kds ddb hudi 25

In a production workload, you can trigger the updates on a schedule or by S3 modification events. A fully automated data lake makes sure your business analysts are always viewing the latest available data.

Clean up the resources

To avoid incurring future charges, follow these steps to remove the example resources:

  1. Delete the resources you created earlier in the pre-requisite section by deleting the stack instances from your stack set, if you created the EMR cluster with the CloudFormation template,.
  2. Stop the cluster via the Amazon EMR console, if you launched the EMR cluster manually.
  3. Empty all the relevant buckets via the Amazon S3 console.

Conclusion

You can build an end-to-end serverless data lake to get real-time insights from DynamoDB by using Kinesis Data Streams—all without writing any complex code. It allows your team to focus on solving business problems by getting useful insights immediately. Application developers have various use cases for moving data quickly through an analytics pipeline, and you can make this happen by enabling Kinesis Data Streams for DynamoDB.

If this post helps you or inspires you to solve a problem, we would love to hear about it! The code for this solution is available in the GitHub repository for you to use and extend. Contributions are always welcome!


About the Authors

dhiraj thakur 100Dhiraj Thakur is a Solutions Architect with Amazon Web Services. He works with AWS customers and partners to guide enterprise cloud adoption, migration, and strategy. He is passionate about technology and enjoys building and experimenting in the analytics and AI/ML space.

 

 

 

Saurabh Shrivastava 100Saurabh Shrivastava is a solutions architect leader and analytics/ML specialist working with global systems integrators. He works with AWS Partners and customers to provide them with architectural guidance for building scalable architecture in hybrid and AWS environments. He enjoys spending time with his family outdoors and traveling to new destinations to discover new cultures.

 

 

 

Dylan QuDylan Qu is an AWS solutions architect responsible for providing architectural guidance across the full AWS stack with a focus on data analytics, AI/ML, and DevOps.