Magalu: 113 applications migrated before Black Friday 2018

01 · CONTEXT

An overloaded back end behind the APIs

According to a case study published by Google Cloud, Magazine Luiza already had, in 2013, an e-commerce platform and an API library. The problem was that these APIs connected to an overloaded back-end application, built with 150,000 lines of code.

1. In the diagram, clients call the API library.

2. The APIs connected to a back-end application.

3. The back end is described as overloaded.

The study adds that deploying the new APIs was slow and that the company faced undesirable dependencies, scalability challenges, and responsibilities spread across isolated teams.

02 · DECOUPLING

A small team and an API platform

Also according to the study, the then operations director assigned a small team of developers, detached from IT management processes and organizational obstacles, with access to all of the company's e-commerce operations.

That team started using the Apigee API management platform, which helped decouple the front end from the back-end systems and made it easier to iterate on the new applications, while other teams kept the legacy systems of record.

1. The new applications sit at the front-end edge.

2. The API platform sits between the front end and the back-end systems and helps decouple them.

3. The legacy systems of record continue to be maintained by other teams.

The summary of a talk on InfoQ Brasil tells a similar story from the organizational side: Luizalabs went from a department focused on research and development to Magazine Luiza's technology and innovation company, motivated by helping the retailer through a technological and cultural transformation.

03 · MIGRATION

113 applications before Black Friday

According to the study, after GCP launched its region in Brazil, Magazine Luiza began deploying workloads on it. The company already used Kubernetes, which was useful for the multi-cloud strategy, and the data migration went smoothly because they used Kubernetes together with their own open-source PaaS.

To improve performance and stability in the face of Black Friday's traffic increase, the company decided to migrate 113 applications to GCP in less than 60 days before that date.

1. In the diagram, the client represents all the platforms served.

2. The API platform gives visibility into API and application activity.

3. The migrated applications now run on Google Cloud.

The diagram combines two pieces of the case study, the API platform and the migrated applications, without stating the exact path of each request.

The reported result is that Black Friday 2018 was the biggest in the company's history, with API traffic levels well above before. The study says Apigee helped with execution, meeting customer demand across all platforms.

04 · SCALE

Catalog, sellers, and small teams

The study compares the current catalog with that of the legacy sales and distribution system and reports the reach of the seller ecosystem. The comparison below uses the numbers as the vendor presents them.

Other merchants join this ecosystem through the API platform. The study also says that, with the ease of creating services, teams can work in groups of five or six people who handle segments of an application, and that Firebase helped reduce the complexity of building the applications that reach customers.

05 · LESSONS

What to take into a design

This case does not state how the gateway was sized, which applications were migrated first, or how traffic was measured. What it supports is the combination of an API layer between front and back, small teams, and a migration on a tight deadline.