Coupang SCM Workflow: Building an Efficient, Scalable Low-Code, No-Code Platform

In an environment where operations teams, developers, BAs, and data scientists work together to create data and feed it back into systems, Coupang's SCM team built the SCM Workflow platform to reduce repetitive requests and development bottlenecks.

The core consists of two parts.

• No-code data builder: Not only developers but also non-developers can access shared data sources such as Redshift, Hive, Presto, Aurora, MySQL, Elasticsearch, and S3 to query, extract, export, and integrate with systems.
• Low-code service builder: Repetitive data processing, delivery, and integration logic is bundled into shared and custom components, allowing developers to focus only on core algorithms when building services.

The data builder can pull product inventory and order data from multiple sources, connect computation nodes, and split queries for parallel processing when needed. Results can be delivered to other teams or systems via workflow ID, URL, file path, and more, and callback and schedule features automate recurring tasks.

The service builder embeds infrastructure features that are repeatedly needed into the platform. It provides versioning, configuration, an error log viewer, alarms, and dashboards to enable debugging and operations from the web, as well as shared blocks such as notification, REST callback, retry, file upload/download/copy, queue message send/receive, and AWS integration.

For scalability and stability, the architecture is divided into 3 layers, each designed to support clustering and scale out. The workflow engine manages models and instances, while actual tasks are executed asynchronously and independently by workers, so that delays or excessive resource use in a specific task have less impact on the overall system.

In the web interface, users can drag and drop blocks in a canvas editor to build models and visually compose execution order and dependencies. It also supports checking execution history and error logs, along with team-based group management and permission control, making it easy for users unfamiliar with coding to build internal applications.

Ultimately, this platform aims to reduce the complex integration and operational burden of existing tools like Jenkins, Airflow, and Notebook, while unifying everything from data creation to visualization and service launch into a single flow.