This can be a visitor submit co-authored by Michael Davies from Open Universities Australia.
At Open Universities Australia (OUA), we empower college students to discover an enormous array of levels from famend Australian universities, all delivered via on-line studying. We provide college students different pathways to attain their academic aspirations, offering them with the pliability and accessibility to succeed in their tutorial targets. Since our founding in 1993, now we have supported over 500,000 college students to attain their targets by offering pathways to over 2,600 topics at 25 universities throughout Australia.
As a not-for-profit group, price is an important consideration for OUA. Whereas reviewing our contract for the third-party device we had been utilizing for our extract, remodel, and cargo (ETL) pipelines, we realized that we may replicate a lot of the identical performance utilizing Amazon Internet Companies (AWS) companies reminiscent of AWS Glue, Amazon AppFlow, and AWS Step Features. We additionally acknowledged that we may consolidate our supply code (a lot of which was saved within the ETL device itself) right into a code repository that may very well be deployed utilizing the AWS Cloud Growth Equipment (AWS CDK). By doing so, we had a possibility to not solely cut back prices but in addition to boost the visibility and maintainability of our knowledge pipelines.
On this submit, we present you ways we used AWS companies to interchange our present third-party ETL device, bettering the crew’s productiveness and producing a big discount in our ETL operational prices.
Our method
The migration initiative consisted of two predominant components: constructing the brand new structure and migrating knowledge pipelines from the prevailing device to the brand new structure. Typically, we’d work on each in parallel, testing one element of the structure whereas creating one other on the identical time.
From early in our migration journey, we started to outline just a few guiding ideas that we’d apply all through the event course of. These had been:
- Easy and modular – Use easy, reusable design patterns with as few shifting components as potential. Construction the code base to prioritize ease of use for builders.
- Price-effective – Use assets in an environment friendly, cost-effective manner. Purpose to reduce conditions the place assets are working idly whereas ready for different processes to be accomplished.
- Enterprise continuity – As a lot as potential, make use of present code quite than reinventing the wheel. Roll out updates in phases to reduce potential disruption to present enterprise processes.
Structure overview
The next Diagram 1 is the high-level structure for the answer.
Diagram 1: General structure of the answer, utilizing AWS Step Features, Amazon Redshift and Amazon S3
The next AWS companies had been used to form our new ETL structure:
- Amazon Redshift – A completely managed, petabyte-scale knowledge warehouse service within the cloud. Amazon Redshift served as our central knowledge repository, the place we’d retailer knowledge, apply transformations, and make knowledge out there to be used in analytics and enterprise intelligence (BI). Notice: The provisioned cluster itself was deployed individually from the ETL structure and remained unchanged all through the migration course of.
- AWS Cloud Growth Equipment (AWS CDK) – The AWS Cloud Growth Equipment (AWS CDK) is an open-source software program growth framework for outlining cloud infrastructure in code and provisioning it via AWS CloudFormation. Our infrastructure was outlined as code utilizing the AWS CDK. Because of this, we simplified the way in which we outlined the assets we wished to deploy whereas utilizing our most popular coding language for growth.
- AWS Step Features – With AWS Step Features, you possibly can create workflows, additionally known as State machines, to construct distributed purposes, automate processes, orchestrate microservices, and create knowledge and machine studying pipelines. AWS Step Features can name over 200 AWS companies together with AWS Glue, AWS Lambda, and Amazon Redshift. We used the AWS Step Perform state machines to outline, orchestrate, and execute our knowledge pipelines.
- Amazon EventBridge – We used Amazon EventBridge, the serverless occasion bus service, to outline the event-based guidelines and schedules that may set off our AWS Step Features state machines.
- AWS Glue – An information integration service, AWS Glue consolidates main knowledge integration capabilities right into a single service. These embody knowledge discovery, trendy ETL, cleaning, remodeling, and centralized cataloging. It’s additionally serverless, which implies there’s no infrastructure to handle. consists of the power to run Python scripts. We used it for executing long-running scripts, reminiscent of for ingesting knowledge from an exterior API.
- AWS Lambda – AWS Lambda is a extremely scalable, serverless compute service. We used it for executing easy scripts, reminiscent of for parsing a single textual content file.
- Amazon AppFlow – Amazon AppFlow allows easy integration with software program as a service (SaaS) purposes. We used it to outline flows that may periodically load knowledge from chosen operational methods into our knowledge warehouse.
- Amazon Easy Storage Service (Amazon S3) – An object storage service providing industry-leading scalability, knowledge availability, safety, and efficiency. Amazon S3 served as our staging space, the place we’d retailer uncooked knowledge previous to loading it into different companies reminiscent of Amazon Redshift. We additionally used it as a repository for storing code that may very well be retrieved and utilized by different companies.
The place sensible, we made use of the file construction of our code base for outlining assets. We arrange our AWS CDK to check with the contents of a particular listing and outline a useful resource (for instance, an AWS Step Features state machine or an AWS Glue job) for every file it present in that listing. We additionally made use of configuration information so we may customise the attributes of particular assets as required.
Particulars on particular patterns
Within the above structure Diagram 1, we confirmed a number of flows by which knowledge may very well be ingested or unloaded from our Amazon Redshift knowledge warehouse. On this part, we spotlight 4 particular patterns in additional element which had been utilized within the remaining resolution.
Sample 1: Knowledge transformation, load, and unload
A number of of our knowledge pipelines included important knowledge transformation steps, which had been primarily carried out via SQL statements executed by Amazon Redshift. Others required ingestion or unloading of information from the information warehouse, which may very well be carried out effectively utilizing COPY or UNLOAD statements executed by Amazon Redshift.
In step with our goal of utilizing assets effectively, we sought to keep away from working these statements from inside the context of an AWS Glue job or AWS Lambda perform as a result of these processes would stay idle whereas ready for the SQL assertion to be accomplished. As an alternative, we opted for an method the place SQL execution duties can be orchestrated by an AWS Step Features state machine, which might ship the statements to Amazon Redshift and periodically verify their progress earlier than marking them as both profitable or failed. The next Diagram 2 exhibits this workflow.
Diagram 2: Knowledge transformation, load, and unload sample utilizing Amazon Lambda and Amazon Redshift inside an AWS Step Perform
Sample 2: Knowledge replication utilizing AWS Glue
In circumstances the place we would have liked to duplicate knowledge from a third-party supply, we used AWS Glue to run a script that may question the related API, parse the response, and retailer the related knowledge in Amazon S3. From right here, we used Amazon Redshift to ingest the information utilizing a COPY assertion. The next Diagram 3 exhibits this workflow.
Diagram 3: Copying from exterior API to Redshift with AWS Glue
Notice: Another choice for this step can be to make use of Amazon Redshift auto-copy, however this wasn’t out there at time of growth.
Sample 3: Knowledge replication utilizing Amazon AppFlow
For sure purposes, we had been ready to make use of Amazon AppFlow flows rather than AWS Glue jobs. Because of this, we may summary a few of the complexity of querying exterior APIs straight. We configured our Amazon AppFlow flows to retailer the output knowledge in Amazon S3, then used an EventBridge rule based mostly on an Finish Stream Run Report occasion (which is an occasion which is revealed when a movement run is full) to set off a load into Amazon Redshift utilizing a COPY assertion. The next Diagram 4 exhibits this workflow.
Through the use of Amazon S3 as an intermediate knowledge retailer, we gave ourselves better management over how the information was processed when it was loaded into Amazon Redshift, in comparison with loading the information on to the information warehouse utilizing Amazon AppFlow.
Diagram 4: Utilizing Amazon AppFlow to combine exterior knowledge to Amazon S3 and replica to Amazon Redshift
Sample 4: Reverse ETL
Though most of our workflows contain knowledge being introduced into the information warehouse from exterior sources, in some circumstances we would have liked the information to be exported to exterior methods as an alternative. This fashion, we may run SQL queries with advanced logic drawing on a number of knowledge sources and use this logic to help operational necessities, reminiscent of figuring out which teams of scholars ought to obtain particular communications.
On this movement, proven within the following Diagram 5, we begin by working an UNLOAD assertion in Amazon Redshift to unload the related knowledge to information in Amazon S3. From right here, every file is processed by an AWS Lambda perform, which performs any mandatory transformations and sends the information to the exterior software via a number of API calls.
Diagram 5: Reverse ETL workflow, sending knowledge again out to exterior knowledge sources
Outcomes
The re-architecture and migration course of took 5 months to finish, from the preliminary idea to the profitable decommissioning of the earlier third-party device. A lot of the architectural effort was accomplished by a single full-time worker, with others on the crew primarily aiding with the migration of pipelines to the brand new structure.
We achieved important price reductions, with remaining bills on AWS native companies representing solely a small proportion of projected prices in comparison with persevering with with the third-party ETL device. Shifting to a code-based method additionally gave us better visibility of our pipelines and made the method of sustaining them faster and simpler. General, the transition was seamless for our finish customers, who had been in a position to view the identical knowledge and dashboards each throughout and after the migration, with minimal disruption alongside the way in which.
Conclusion
Through the use of the scalability and cost-effectiveness of AWS companies, we had been in a position to optimize our knowledge pipelines, cut back our operational prices, and enhance our agility.
Pete Allen, an analytics engineer from Open Universities Australia, says, “Modernizing our knowledge structure with AWS has been transformative. Transitioning from an exterior platform to an in-house, code-based analytics stack has vastly improved our scalability, flexibility, and efficiency. With AWS, we will now course of and analyze knowledge with a lot quicker turnaround, decrease prices, and better availability, enabling speedy growth and deployment of information options, resulting in deeper insights and higher enterprise selections.”
Extra assets
Concerning the Authors
Michael Davies is a Knowledge Engineer at OUA. He has in depth expertise inside the training {industry}, with a specific give attention to constructing strong and environment friendly knowledge structure and pipelines.
Emma Arrigo is a Options Architect at AWS, specializing in training prospects throughout Australia. She makes a speciality of leveraging cloud know-how and machine studying to handle advanced enterprise challenges within the training sector. Emma’s ardour for knowledge extends past her skilled life, as evidenced by her canine named Knowledge.
