Optimizing Frontend‑Backend Collaboration with Data Direct Access at Baidu Commercial Frontend Team

Introduction

In most scenarios, front‑end and back‑end can agree on data interfaces before development, allowing parallel work, but large systems often have intermediate modules that hinder direct data access, leading to inefficient environment management and reduced developer experience.

1. Practice Plan

The team focused on two goals: collaboration efficiency and quality/experience optimization. Collaboration efficiency includes building basic capabilities and upgrading collaboration models, while quality assurance ensures that data used in development matches the logic that will be released.

2. Data Direct Capability

The backend maintains a BFF layer that adapts upstream data for the front‑end. By pulling offline material data from Redis and associating it with device identifiers, the backend can replace online requests with offline data, allowing the front‑end to work against a constantly updated, stable environment without managing multiple test setups.

3. Upgrade Collaboration Mode

Using data‑direct capability, the team introduced stub services: when a request carries a special marker, the backend skips normal processing and returns stub data, enabling parallel development before backend code is deployed.

4. Data Tiering

Data usage is classified into three types: manually generated, backend‑generated offline, and backend‑generated online. Early development uses editable stub data, integration uses code‑generated data with enforced realism, and post‑deployment uses live data collected from production, ensuring both efficiency and correctness.

5. Platform Experience Optimization

To handle large projects with hundreds of data items, the team introduced a "fragment" concept for batch editing, allowing a single edit operation to be applied to multiple data entries and supporting versioned fragments for compatibility testing.

6. Summary

The optimized front‑end/back‑end collaboration reduced average project delivery time by more than 50%, with thousands of business projects successfully using the approach, and the team continues to explore extensions to visual acceptance and other scenarios.