It is a visitor publish co-authored by Michael Davies from Open Universities Australia.
At Open Universities Australia (OUA), we empower college students to discover an unlimited array of levels from famend Australian universities, all delivered by means of on-line studying. We provide college students various pathways to realize their academic aspirations, offering them with the pliability and accessibility to achieve their educational targets. Since our founding in 1993, we’ve got supported over 500,000 college students to realize 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 software 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 Providers (AWS) providers similar to AWS Glue, Amazon AppFlow, and AWS Step Capabilities. We additionally acknowledged that we may consolidate our supply code (a lot of which was saved within the ETL software itself) right into a code repository that may very well be deployed utilizing the AWS Cloud Improvement Package (AWS CDK). By doing so, we had a possibility to not solely cut back prices but additionally to boost the visibility and maintainability of our knowledge pipelines.
On this publish, we present you ways we used AWS providers to interchange our current third-party ETL software, bettering the group’s productiveness and producing a major discount in our ETL operational prices.
Our strategy
The migration initiative consisted of two essential components: constructing the brand new structure and migrating knowledge pipelines from the present software to the brand new structure. Usually, we’d work on each in parallel, testing one part of the structure whereas creating one other on the identical time.
From early in our migration journey, we started to outline a number of 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 doable. Construction the code base to prioritize ease of use for builders.
- Price-effective – Use assets in an environment friendly, cost-effective method. Intention to attenuate conditions the place assets are operating idly whereas ready for different processes to be accomplished.
- Enterprise continuity – As a lot as doable, make use of current code somewhat than reinventing the wheel. Roll out updates in phases to attenuate potential disruption to current 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 Capabilities, Amazon Redshift and Amazon S3
The next AWS providers 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 obtainable 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 Improvement Package (AWS CDK) – The AWS Cloud Improvement Package (AWS CDK) is an open-source software program improvement framework for outlining cloud infrastructure in code and provisioning it by means of AWS CloudFormation. Our infrastructure was outlined as code utilizing the AWS CDK. Consequently, we simplified the way in which we outlined the assets we wished to deploy whereas utilizing our most popular coding language for improvement.
- AWS Step Capabilities – With AWS Step Capabilities, you’ll be able to create workflows, additionally referred to as State machines, to construct distributed purposes, automate processes, orchestrate microservices, and create knowledge and machine studying pipelines. AWS Step Capabilities can name over 200 AWS providers 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 might set off our AWS Step Capabilities state machines.
- AWS Glue – An information integration service, AWS Glue consolidates main knowledge integration capabilities right into a single service. These embody knowledge discovery, fashionable ETL, cleaning, reworking, and centralized cataloging. It’s additionally serverless, which suggests there’s no infrastructure to handle. contains the flexibility to run Python scripts. We used it for executing long-running scripts, similar to 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, similar to 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 might periodically load knowledge from chosen operational programs 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 providers similar to Amazon Redshift. We additionally used it as a repository for storing code that may very well be retrieved and utilized by different providers.
The place sensible, we made use of the file construction of our code base for outlining assets. We arrange our AWS CDK to discuss with the contents of a particular listing and outline a useful resource (for instance, an AWS Step Capabilities 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 answer.
Sample 1: Knowledge transformation, load, and unload
A number of of our knowledge pipelines included vital knowledge transformation steps, which had been primarily carried out by means of SQL statements executed by Amazon Redshift. Others required ingestion or unloading of information from the info 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 operating these statements from throughout 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 strategy the place SQL execution duties can be orchestrated by an AWS Step Capabilities 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 reveals 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 wanted to copy knowledge from a third-party supply, we used AWS Glue to run a script that might 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 info utilizing a COPY assertion. The next Diagram 3 reveals 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 obtainable at time of improvement.
Sample 3: Knowledge replication utilizing Amazon AppFlow
For sure purposes, we had been ready to make use of Amazon AppFlow flows instead of AWS Glue jobs. Consequently, we may summary among the complexity of querying exterior APIs immediately. 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 printed when a move run is full) to set off a load into Amazon Redshift utilizing a COPY assertion. The next Diagram 4 reveals this workflow.
Through the use of Amazon S3 as an intermediate knowledge retailer, we gave ourselves better management over how the info was processed when it was loaded into Amazon Redshift, in comparison with loading the info on to the info warehouse utilizing Amazon AppFlow.
Diagram 4: Utilizing Amazon AppFlow to combine exterior knowledge to Amazon S3 and duplicate to Amazon Redshift
Sample 4: Reverse ETL
Though most of our workflows contain knowledge being introduced into the info warehouse from exterior sources, in some circumstances we wanted the info to be exported to exterior programs as an alternative. This manner, we may run SQL queries with advanced logic drawing on a number of knowledge sources and use this logic to assist operational necessities, similar to figuring out which teams of scholars ought to obtain particular communications.
On this move, proven within the following Diagram 5, we begin by operating 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 essential transformations and sends the info to the exterior utility by means of 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 software. Many of the architectural effort was accomplished by a single full-time worker, with others on the group primarily aiding with the migration of pipelines to the brand new structure.
We achieved vital price reductions, with remaining bills on AWS native providers representing solely a small share of projected prices in comparison with persevering with with the third-party ETL software. Transferring to a code-based strategy 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 providers, 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 are able to now course of and analyze knowledge with a lot sooner turnaround, decrease prices, and better availability, enabling fast improvement and deployment of information options, resulting in deeper insights and higher enterprise selections.”
Further assets
In regards to the Authors
Michael Davies is a Knowledge Engineer at OUA. He has in depth expertise throughout the training {industry}, with a selected concentrate on constructing sturdy and environment friendly knowledge structure and pipelines.
Emma Arrigo is a Options Architect at AWS, specializing in training clients 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.
