Convert Oracle XML BLOB information to JSON utilizing Amazon EMR and cargo to Amazon Redshift

In legacy relational database administration methods, information is saved in a number of complicated information varieties, such XML, JSON, BLOB, or CLOB. This information would possibly comprise priceless info that’s typically troublesome to rework into insights, so that you is perhaps on the lookout for methods to load and use this information in a contemporary cloud information warehouse comparable to Amazon Redshift. One such instance is migrating information from a legacy Oracle database with XML BLOB fields to Amazon Redshift, by performing preprocessing and conversion of XML to JSON utilizing Amazon EMR. On this publish, we describe an answer structure for this use case, and present you the way to implement the code to deal with the XML conversion.

Answer overview

Step one in any information migration challenge is to seize and ingest the information from the supply database. For this job, we use AWS Database Migration Service (AWS DMS), a service that helps you migrate databases to AWS rapidly and securely. On this instance, we use AWS DMS to extract information from an Oracle database with XML BLOB fields and stage the identical information in Amazon Easy Storage Service (Amazon S3) in Apache Parquet format. Amazon S3 is an object storage service providing industry-leading scalability, information availability, safety, and efficiency, and is the storage of alternative for organising information lakes on AWS.

After the information is ingested into an S3 staging bucket, we used Amazon EMR to run a Spark job to carry out the conversion of XML fields to JSON fields, and the outcomes are loaded in a curated S3 bucket. Amazon EMR runtime for Apache Spark might be over thrice quicker than clusters with out EMR runtime, and has 100% API compatibility with normal Apache Spark. This improved efficiency means your workloads run quicker and it saves you compute prices, with out making any adjustments to your utility.

Lastly, reworked and curated information is loaded into Amazon Redshift tables utilizing the COPY command. The Amazon Redshift desk construction ought to match the variety of columns and the column information varieties within the supply file. As a result of we saved the information as a Parquet file, we specify the SERIALIZETOJSON choice within the COPY command. This enables us to load complicated varieties, comparable to construction and array, in a column outlined as SUPER information sort within the desk.

The next structure diagram reveals the end-to-end workflow.

Intimately, AWS DMS migrates information from the supply database tables into Amazon S3, in Parquet format. Apache Spark on Amazon EMR reads the uncooked information, transforms the XML information sort into JSON, and saves the information to the curated S3 bucket. In our code, we used an open-source library, known as spark-xml, to parse and question the XML information.

In the remainder of this publish, we assume that the AWS DMS duties have already run and created the supply Parquet recordsdata within the S3 staging bucket. If you wish to arrange AWS DMS to learn from an Oracle database with LOB fields, check with Successfully migrating LOB information to Amazon S3 from Amazon RDS for Oracle with AWS DMS or watch the video Migrate Oracle to S3 Information lake by way of AWS DMS.


If you wish to comply with together with the examples on this publish utilizing your AWS account, we offer an AWS CloudFormation template you may launch by selecting Launch Stack:


Present a stack identify and depart the default settings for every part else. Look forward to the stack to show Create Full (this could solely take a couple of minutes) earlier than transferring on to the opposite sections.

The template creates the next assets:

  • A digital personal cloud (VPC) with two personal subnets which have routes to an Amazon S3 VPC endpoint
  • The S3 bucket {stackname}-s3bucket-{xxx}, which accommodates the next folders:
    • libs – Comprises the JAR file so as to add to the pocket book
    • notebooks – Comprises the pocket book to interactively check the code
    • information – Comprises the pattern information
  • An Amazon Redshift cluster, in one of many two personal subnets, with a database named rs_xml_db and a schema named rs_xml
  • A secret (rs_xml_db) in AWS Secrets and techniques Supervisor
  • An EMR cluster

The CloudFormation template shared on this publish is solely for demonstration functions solely. Please conduct your individual safety evaluate and incorporate finest practices previous to any manufacturing deployment utilizing artifacts from the publish.

Lastly, some fundamental data of Python and Spark DataFrames may also help you evaluate the transformation code, however isn’t obligatory to finish the instance.

Understanding the pattern information

On this publish, we use faculty college students’ course and topics pattern information that we created. Within the supply system, information consists of flat construction fields, like course_id and course_name, and an XML discipline that features all of the course materials and topics concerned within the respective course. The next screenshot is an instance of the supply information, which is staged in an S3 bucket as a prerequisite step.

We will observe that the column study_material_info is an XML sort discipline and accommodates nested XML tags in it. Let’s see the way to convert this nested XML discipline to JSON within the subsequent steps.

Run a Spark job in Amazon EMR to rework the XML fields within the uncooked information to JSON

On this step, we use an Amazon EMR pocket book, which is a managed atmosphere to create and open Jupyter Pocket book and JupyterLab interfaces. It lets you interactively analyze and visualize information, collaborate with friends, and construct purposes utilizing Apache Spark on EMR clusters. To open the pocket book, comply with these steps:

  1. On the Amazon S3 console, navigate to the bucket you created as a prerequisite step.
  2. Obtain the file within the notebooks folder.
  3. On the Amazon EMR console, select Notebooks within the navigation pane.
  4. Select Create pocket book.
  5. For Pocket book identify, enter a reputation.
  6. For Cluster, choose Select an present cluster.
  7. Choose the cluster you created as a prerequisite.
  8. For Safety Teams, select BDB1909-EMR-LIVY-SG and BDB1909-EMR-Pocket book-SG
  9. For AWS Service Position, select the position bdb1909-emrNotebookRole-{xxx}.
  10. For Pocket book location, specify the S3 path within the notebooks folder (s3://{stackname}-s3bucket-xxx}/notebooks/).
  11. Select Create pocket book.
  12. When the pocket book is created, select Open in JupyterLab.
  13. Add the file you downloaded earlier.
  14. Open the brand new pocket book.

    The pocket book ought to look as proven within the following screenshot, and it accommodates a script written in Scala.
  15. Run the primary two cells to configure Apache Spark with the open-source spark-xml library and import the wanted modules.The spark-xml package deal permits studying XML recordsdata in native or distributed file methods as Spark DataFrames. Though primarily used to transform (parts of) giant XML paperwork right into a DataFrame, spark-xml may also parse XML in a string-valued column in an present DataFrame with the from_xml operate, with the intention to add it as a brand new column with parsed outcomes as a struct.
  16. To take action, within the third cell, we load the information from the Parquet file generated by AWS DMS right into a DataFrame, then we extract the attribute that accommodates the XML code (STUDY_MATERIAL_INFO) and map it to a string variable identify payloadSchema.
  17. We will now use the payloadSchema within the from_xml operate to transform the sector STUDY_MATERIAL_INFO right into a struct information sort and added it as a column named course_material in a brand new DataFrame parsed.
  18. Lastly, we will drop the unique discipline and write the parsed DataFrame to our curated zone in Amazon S3.

Because of the construction variations between DataFrame and XML, there are some conversion guidelines from XML information to DataFrame and from DataFrame to XML information. Extra particulars and documentation can be found XML Information Supply for Apache Spark.

Once we convert from XML to DataFrame, attributes are transformed as fields with the heading prefix attributePrefix (underscore (_) is the default). For instance, see the next code:

  <ebook class="undergraduate">
    <title lang="en">Introduction to Biology</title>
    <creator>Demo Creator 1</creator>

It produces the next schema:

 |-- class: string (nullable = true)
 |-- title: struct (nullable = true)
 |    |-- _VALUE: string (nullable = true)
 |    |-- _lang: string (nullable = true)
 |-- creator: string (nullable = true)
 |-- yr: string (nullable = true)
 |-- value: string (nullable = true)

Subsequent, we’ve a worth in a component that has no little one parts however attributes. The worth is put in a separate discipline, valueTag. See the next code:

<title lang="en">Introduction to Biology</title>

It produces the next schema, and the tag lang is transformed into the _lang discipline contained in the DataFrame:

|-- title: struct (nullable = true)
 |    |-- _VALUE: string (nullable = true)
 |    |-- _lang: string (nullable = true)

Copy curated information into Amazon Redshift and question tables seamlessly

As a result of our semi-structured nested dataset is already written within the S3 bucket as Apache Parquet formatted recordsdata, we will use the COPY command with the SERIALIZETOJSON choice to ingest information into Amazon Redshift. The Amazon Redshift desk construction ought to match the metadata of the Parquet recordsdata. Amazon Redshift can exchange any Parquet columns, together with construction and array varieties, with SUPER information columns.

The next code demonstrates CREATE TABLE instance to create a staging desk.

create desk rs_xml_db.public.stg_edw_course_catalog 
course_id bigint,
course_name character various(5000),
course_material tremendous

The next code makes use of the COPY instance to load from Parquet format:

COPY rs_xml_db.public.stg_edw_course_catalog FROM 's3://<<your Amazon S3 Bucket for curated information>>/information/goal/<<your output parquet file>>' 
IAM_ROLE '<<your IAM position>>' 

By utilizing semistructured information assist in Amazon Redshift, you may ingest and retailer semistructured information in your Amazon Redshift information warehouses. With the SUPER information sort and PartiQL language, Amazon Redshift expands the information warehouse functionality to combine with each SQL and NoSQL information sources. The SUPER information sort solely helps as much as 1 MB of information for a person SUPER discipline or object. Observe, the JSON object could also be saved in a SUPER information sort, however studying this information utilizing JSON capabilities presently has a VARCHAR (65535 byte) restrict. See Limitations for extra particulars.

The next instance reveals how nested JSON might be simply accessed utilizing SELECT statements:

SELECT DISTINCT bk._category
FROM rs_xml_db.public.stg_edw_course_catalog most important
INNER JOIN most important.course_material.ebook bk ON true;

The next screenshot reveals our outcomes.

Clear up

To keep away from incurring future costs, first delete the pocket book and the associated recordsdata on Amazon S3 bucket as defined in this EMR documentation web page then the CloudFormation stack.


This publish demonstrated the way to use AWS providers like AWS DMS, Amazon S3, Amazon EMR, and Amazon Redshift to seamlessly work with complicated information varieties like XML and carry out historic migrations when constructing a cloud information lake home on AWS. We encourage you to do this resolution and reap the benefits of all the advantages of those purpose-built providers.

In case you have questions or strategies, please depart a remark.

In regards to the authors

Abhilash Nagilla is a Sr. Specialist Options Architect at AWS, serving to public sector prospects on their cloud journey with a concentrate on AWS analytics providers. Outdoors of labor, Abhilash enjoys studying new applied sciences, watching films, and visiting new locations.

Avinash Makey is a Specialist Options Architect at AWS. He helps prospects with information and analytics options in AWS. Outdoors of labor he performs cricket, tennis and volleyball in free time.

Fabrizio Napolitano is a Senior Specialist SA for DB and Analytics. He has labored within the analytics area for the final 20 years, and has not too long ago and fairly abruptly turn out to be a Hockey Dad after transferring to Canada.

Latest articles

Related articles

Leave a reply

Please enter your comment!
Please enter your name here