Design and Implementation of Full‑Link Load Testing at Dada Group

Dada Group, a Shanghai‑based instant‑delivery platform, needed a reliable way to ensure system stability as order volume grew beyond 10 million daily deliveries, prompting the development of a full‑link load‑testing solution.

Industry vs. Dada approaches : Traditional traffic‑labeling (shadow DB/Cache/Queue) isolates test traffic but requires extensive code changes across heterogeneous services, which Dada rejected. Instead, Dada adopted a machine‑tagging strategy, abstracting each DB, Redis, and ES instance as a node and registering them in a service registry.

Core component – Link Governance SDK : The SDK routes requests based on link type (benchmark or production), enabling separate test and production flows at the machine level.

Testing platform : Built on a customized JMeter core, the platform includes a front‑end UI, task parser, execution engine, and result processor, providing real‑time visual dashboards and dynamic reporting.

Implementation steps : (1) Abstract machines into nodes; (2) Register node info; (3) Integrate services with the Link Governance SDK; (4) Launch services, retrieve node data, and establish DB/Redis/MQ connections.

Full‑link rollout : Divided into pre‑test (link mapping, optimization plans, fine‑grained models), test (validation, warm‑up, execution, metric monitoring), and post‑test (reporting, issue tracing, capacity estimation, retrospectives).

Optimization plans cover thread‑pool/connection‑pool scaling, MySQL binlog tuning, Redis bandwidth expansion, and MQ consumer scaling, with specific MySQL parameters highlighted.

Fine‑grained models consist of data models (shadowed rider, merchant, order data) and traffic models (order dispatch flow), using manually constructed traffic that reflects time‑ and space‑based peak patterns.

Pre‑heat strategy introduces hot data into the test environment to align response times with production.

Post‑test analysis compares TPS, latency, and middleware metrics between test and production to refine models.

Results : Over four major promotional events since Q1 2019, the solution isolated traffic and data, reduced machine costs by 40%, improved labor efficiency by 65%, and consistently uncovered over ten performance issues per event, enhancing system stability.

Authors : Senior engineers and architects from Dada Group who led the load‑testing initiatives.