How AWS Serverless Has Streamlined a Car Factory’s Supply Chain

In this video, Mr. Lukasz Panusz, the Chief Architect of PGS Software is interviewed on how AWS Serverless has streamlined a car factory’s supply chain.

At 0:20 he begins by stating that PGS software is a Poland based company which specializes in delivering custom-made solutions using the latest cutting-edge technologies for clients.

At 0:37 he explains one of their projects asked was the car manufacturer wanted to completely switch to just-in-time delivery to limit the cost of the warehousing. They were asked to come up with a cost-effective performance and resilient solution that would simplify the full process.

At 1:15 Mr. Lukasz begins to explain the solution. He states that the car manufacturer is producing in the order of about 100-200 files a day, having multiple orders which are then gathered to the OFTP2 server that stores as well as synchronizes to the primary S3 bucket. At 1:43 he states that each of the files contains it’s own unique ID, it is then processed to the splitter which triggers the first Lambda function. At 2:04 he adds that the Lambda function is responsible for validation of business file structures and the order of the content.

At 2:16 he states that based on the inputs, two outcomes are produced where the first one is breaking down of the orders into separate files while the second utilizes the message stores in SQS for further processing and notification.

SQS (Simple Queue Service) then triggers another Lambda function names worker that is responsible for business processing of the order. At 2:53 he states the information is pulled from Redis about the supplier, car parts and once it is successful, they are stored back to S3.

At 4:09 he describes what happens in the failure cases. He emphasizes that all the Lambda functions are implemented according to the reactive manifesto using back pressure.

At 4:30 he highlights that the unknown suppliers are completely stored in a new separate ordering file named unknown and saved into the bucket called erroneous which triggers the retry logic. At 4:53 he adds that the retry logic gets triggered by the cloud watch scheduled events to the repeater, and the notification that takes information form missing data from Redis which are then sent in email form through the SNS.

At 5:23 he adds that the repeater Lambda function takes orders in an unknown state, publish them again to SQS to reuse the container flow.

At 6:02 Mr. Lukasz states that initially the step functions were evaluated for the workflow but it turned out that there will be nearly 120 times more expensive. There are also scenarios where single order can contain up to 114 megabytes of data translated to 200,000 orders, and after a day of processing will produce 22.5 billion of messages to be processed.

At 7:17 he states that this solution was brought up in 4 weeks and will need more optimization. He begins to explain that the connecting part between OFTP2 and S3 events to be replaced to something more resilient and persistent.

At 7:43 he adds on that since raw data are stored and processed, it would be the right entry point for the big data solution that can be further extended for the business. Mr. Lukasze summarizes how AWS Serverless has been able to meet the streamlining needs of the car factory supply chain.