Most organizations generate data in real time and ever-increasing volumes. Data is captured from a variety of sources, such as transactional and reporting databases, application logs, customer-facing websites, and external feeds. Companies want to capture, transform, and analyze this time-sensitive data to improve customer experiences, increase efficiency, and drive innovations. With increased data volume and velocity, it’s imperative to capture the data from source systems as soon as they are generated and store them on a secure, scalable, and cost-efficient platform.
AWS Database Migration Service (AWS DMS) performs continuous data replication using change data capture (CDC). Using CDC, you can determine and track data that has changed and provide it as a stream of changes that a downstream application can consume and act on. Most database management systems manage a transaction log that records changes made to the database contents and metadata. AWS DMS reads the transaction log by using engine-specific API operations and functions and captures the changes made to the database in a nonintrusive manner.
Amazon Simple Storage Service (Amazon S3) is the largest and most performant object storage service for structured and unstructured data and the storage service of choice to build a data lake. With Amazon S3, you can cost-effectively build and scale a data lake of any size in a secure environment where data is protected by 99.999999999% of durability.
AWS DMS offers many options to capture data changes from relational databases and store the data in columnar format (Apache Parquet) into Amazon S3:
- AWS DMS to migrate data into Amazon S3 in Parquet format. For more information, see Announcing the support or Parquet data format in AWS DMS 3.1.3.
- Stream AWS DMS data into Amazon Kinesis Data Streams and convert data into Parquet format with Amazon Kinesis Data Firehose and store into Amazon S3.
The second option helps you build a flexible data pipeline to ingest data into an Amazon S3 data lake from several relational and non-relational data sources, compared to just relational data sources support in the former option. Kinesis Data Firehose provides pre-built AWS Lambda blueprints for converting common data sources such as Apache logs and system logs to JSON and CSV formats or writing your own custom functions. It can also convert the format of incoming data from JSON to Parquet or Apache ORC before storing the data in Amazon S3. Data stored in columnar format gives you faster and lower-cost queries with downstream analytics services like Amazon Athena.
In this post, we focus on the technical challenges outlined in the second option and how to address them.
As shown in the following reference architecture, data is ingested from a database into Parquet format in Amazon S3 via AWS DMS integrating with Kinesis Data Streams and Kinesis Data Firehose.
Our solution provides flexibility to ingest data from several sources using Kinesis Data Streams and Kinesis Data Firehose with built-in data format conversion and integrated data transformation capabilities before storing data in a data lake. For more information about data ingestion into Kinesis Data Streams, see Writing Data into Amazon Kinesis Data Streams. You can then query Parquet data in Amazon S3 efficiently with Athena.
Implementing the architecture
AWS DMS can migrate data to and from most widely used commercial and open-source databases. You can migrate and replicate data directly to Amazon S3 in CSV and Parquet formats, and store data in Amazon S3 in Parquet because it offers efficient compression and encoding schemes. Parquet format allows compression schemes on a per-column level, and is future-proofed to allow adding more encodings as they are invented and implemented.
AWS DMS supports Kinesis Data Streams as a target. Kinesis Data Streams is a massively scalable and durable real-time data streaming service that can collect and process large streams of data records in real time. AWS DMS service publishes records to a data stream using JSON. For more information about configuration details, see Use the AWS Database Migration Service to Stream Change Data to Amazon Kinesis Data Streams.
Kinesis Data Firehose can pull data from Kinesis Data Streams. It’s a fully managed service that delivers real-time streaming data to destinations such as Amazon S3, Amazon Redshift, Amazon Elasticsearch Service (Amazon ES), and Splunk. Kinesis Data Firehose can convert the format of input data from JSON to Parquet or ORC before sending it to Amazon S3. It needs reference schema to interpret the AWS DMS streaming data in JSON and convert into Parquet. In this post, we use AWS Glue, a fully managed ETL service, to create a schema in the AWS Glue Data Catalog for Kinesis Data Firehose to reference.
When AWS DMS migrates records, it creates additional fields (metadata) for each migrated record. The metadata provides additional information about the record being migrated, such as source table name, schema name, and type of operation. Most metadata fields add – in their field names (for example, record-type, schema-name, table-name, transaction-id). See the following code:
Additional metadata added by AWS DMS leads to an error during the data format conversion phase in Kinesis Data Firehose. Kinesis Data Firehose follows Hive style formatting and therefore doesn’t recognize the – character in the metadata field names during data conversion from JSON into Parquet and returns an error message: expected at the position 30 of ‘struct’ but ‘-’ is found. For example, see the following code:
You can resolve the issue by making the following changes: specifying JSON key mappings and creating a reference table in AWS Glue before configuring Kinesis Data Firehose.
Specifying JSON key mappings
In your Kinesis Data Firehose configuration, specify JSON key mappings for fields with – in their names. Mapping transforms these specific metadata fields names to _ (for example, record-type changes to record_type).
Use AWS Command Line Interface (AWS CLI) to create Kinesis Data Firehose with the JSON key mappings. Modify the parameters to meet your specific requirements.
Kinesis Data Firehose configuration mapping is only possible through the AWS CLI or API and not through the AWS Management Console.
The following code configures Kinesis Data Firehose with five columns with – in their field names mapped to new field names with _”:
Creating a reference table in AWS Glue
Because Kinesis Data Firehose uses the Data Catalog to reference schema for Parquet format conversion, you must first create a reference table in AWS Glue before configuring Kinesis Data Firehose. Use Athena to create a Data Catalog table. For instructions, see CREATE TABLE. In the table, make sure that the column name uses _ in their names, and manually modify it in advance through the Edit schema option for the referenced table in AWS Glue, if needed.
Use Athena to query the results of data ingested by Kinesis Data Firehose into Amazon S3.
This solution is only applicable in the following use cases:
- Capturing data changes from your source with AWS DMS
- Converting data into Parquet with Kinesis Data Firehose
If you want to store data in non-Parquet format (such CSV or JSON) or ingest into Kinesis through other routes, then you don’t need to modify your Kinesis Data Firehose configuration.
This post demonstrated how to convert AWS DMS data into Parquet format and specific configurations to make sure metadata follows the expected format of Kinesis Data Streams and Kinesis Data Firehose. We encourage you to try this solution and take advantage of all the benefits of using AWS DMS with Kinesis Data Streams and Kinesis Data Firehose. For more information, see Getting started with AWS Database Migration Service and Setting up Amazon Kinesis Firehose.
If you have questions or suggestions, please leave a comment.
About the Author
Viral Shah is a Data Lab Architect with Amazon Web Services. Viral helps our customers architect and build data and analytics prototypes in just four days in the AWS Data Lab. He has over 20 years of experience working with enterprise customers and startups primarily in the Data and Database space.