JD Logistics’ growing demand for real-time operational analytics exposed limitations in its existing data architecture. Data services were developed independently, query engines were fragmented, and users often needed to retrieve data from multiple systems before performing analysis. Repetitive service development and maintenance also placed substantial demands on engineering teams.
To address these challenges, the Operations Data Products team built Udata, an integrated data services and analytics platform powered by StarRocks. Udata combines a unified query engine with configuration-driven metric development, centralized metric management, and data map capabilities.
The low-code development approach reduced metric development time from one or two days to 30 minutes. It also lowered the barrier to creating data services, enabling users outside engineering teams to configure and publish services based on their needs. Centralized metric management makes existing metrics easier to find, maintain, and reuse.
For cross-source analytics, JD Logistics extended StarRocks’ federated query capabilities. The team implemented aggregate function pushdown for MySQL and Elasticsearch, allowing these engines to aggregate data before returning results instead of transferring detailed records for local aggregation. JSF and HTTP integrations also made existing services accessible through SQL, enabling users to reuse established metrics and combine them with other sources.
StarRocks supports both real-time and offline analytics within Udata. Its Primary Key table model accommodates updates to operational records, while its single-table and multi-table join capabilities support analytical workloads without relying exclusively on denormalized tables.
By integrating analytics and data services, Udata enables users to move directly from discovering useful data to publishing corresponding services. This helps JD Logistics reduce repetitive development, improve data reuse, and shorten the feedback loop between analysis and business applications.
At the time of publication, the team’s roadmap included data lake analytics, broader data source support, cross-cluster federated queries, and further consolidation of real-time storage. These were planned directions rather than completed outcomes.