How JD Global Sales Boosted UI Test Speed by 100× with a Three‑Layer Multilingual Testing Framework

Background and Industry Challenges

Multilingual testing commonly suffers from three problems: missing language keys during material production, lack of standardized translation workflows, and low fidelity of overseas test simulations. These issues cause UI text to disappear or display incorrectly, especially when scaling to new languages.

JD‑Specific Challenges

The JD app is a carrier‑grade application with four dimensions—region/country, language, currency, and time zone—resulting in a combinatorial explosion of test scenarios. Additional difficulties include complex UI layers across more than 20 business domains, handling dozens of currencies with specific display formats, and time‑zone conversions that must respect user language settings.

Technical Stack Complexity

Client‑Side

Pages retrieve "country/region", "language", and "mode" from a multilingual SDK. Two scenarios exist: using JD’s retail network library (which automatically uploads these parameters) or handling them manually.

Server‑Side

Parameters flow: client → network library (or non‑network) → color gateway (or non‑color) → SOA → backend services (JSF implicit passing). Services read the global context SDK and must not tamper with the values. Backend services integrate Dongboot kernel components (donglog, dongcontext, dongthread).

Currency Design

Foreign amount = local amount × exchange rate. VND, KRW, and JPY are rounded to the nearest integer; other currencies keep up to two decimal places, trimming trailing zeros.

Time‑Zone Design

If upstream returns a string containing a date/time, the backend performs time‑zone conversion. If a timestamp/Date object is returned, the SOA layer handles the conversion.

QA Challenges

QA must cover a massive matrix of pages, languages, currencies, and time zones, leading to hundreds of page scenarios and a Cartesian product of variations. Personalized pages (e.g., recommendations, messaging) and checkout flows that require precise SKU and pricing data further increase complexity.

Testing Strategy

Goal

Automate problem detection, accelerate review cycles, and provide remediation suggestions to improve the localized experience.

Overall Approach

Adopt a three‑layer verification model: interface layer, page layer, and user‑flow layer. Combine currency‑rate throttling to test fallback scenarios.

Interface Layer

Identify all price‑calculation APIs, set currency contexts, and run batch tests. For middle‑platform services, set DongContext in JSF interfaces; for front‑end services, configure the color gateway to inject the context. Performance target: 80 ms per call, with graceful degradation if exceeded.

Page Layer

One‑click execution of core scenario combinations with automated result validation. Core scenarios include:

HK‑Traditional‑HKD‑UTC+8

TW‑English‑TWD‑UTC+8

SG‑English‑SGD‑UTC+8

AU‑English‑AUD‑UTC+11

CN‑English‑USD‑UTC+8

Automation handles language switching, screenshot capture, navigation, and OCR‑based verification.

Continuous Regression

Integrate automated regression into release pipelines, schedule periodic runs, and generate reports showing pass rates, error counts, and overall coverage.

Intelligent Agent

Leverage a cyber‑platform AI agent to extract translated strings from pages, feed them to a GPT‑based MQM model, and assess translation quality automatically.

Efficiency Gains

Manual UI testing of 36 pages across nine locale combinations required 3 240 minutes; automation reduced this to ~30 minutes (>100× speedup). API batch testing cut execution time by over 70 %.

Future Plans

Expand real‑device cloud coverage for diverse brands, models, and resolutions.

Automate bug filing and module‑level health checks.

Embed large‑model translation quality metrics into the development workflow.