Official Didi technology account
Didi’s unified real‑time data stack outlines best‑practice component choices for four key scenarios—metric monitoring, BI analysis, online services, and feature/tag systems—detailing pipelines from source to sink, resource‑usage guidelines, and a one‑stop development platform to build stable, high‑performance streaming solutions.
Didi balances ride‑hailing supply and demand by forecasting regional needs with time‑series and deep‑learning models, then optimally repositioning drivers through integer programming and refining policies via imitation and offline reinforcement learning, ultimately enhancing passenger experience and platform efficiency.
Laser (Latent Surrogate Representation learning) estimates long‑term ride‑hailing market effects by inferring hidden surrogate variables from short‑term outcomes using an iVAE and inverse‑probability weighting, thereby reducing experiment cost and latency while achieving more accurate causal effect predictions than existing baselines.
After five years of open-source development, Didi’s Mpx framework now powers all internal mini-programs and, thanks to recent upgrades such as composition-API standards, asynchronous sub-package builds, and built-in unit testing, offers a unified, high-performance solution for cross-platform mini-program and web development.
Didi’s redesigned Spatial‑Temporal Supply‑Demand System replaces a single‑Redis bottleneck with a multi‑cluster routing layer, semantic sharding, multi‑level caching and delayed queues, achieving higher horizontal scalability, fault isolation, ~30 % latency reduction, increased cache hit rates, fewer query nodes, and faster, code‑free feature configuration.
The article surveys Didi’s driver‑passenger matching challenges and presents a suite of solutions—from greedy nearest‑driver and Kuhn‑Munkres bipartite matching to stable marriage, dynamic and one‑to‑many assignments, reinforcement‑learning, routing and queueing models—while validating assumptions statistically, integrating preference‑aware machine learning, and outlining multi‑objective and digital‑twin future research.
The article introduces Didi’s large‑scale ride‑hailing marketplace, explaining how mechanism design, decision intelligence, operations research, reinforcement learning, and causal inference are combined to create fair, efficient, data‑driven matching, routing, and incentive systems that tackle massive two‑sided market challenges.
DiDi’s Omega platform provides an end‑to‑end traffic‑data pipeline—from SDK collection through real‑time and offline ETL to storage and analysis—augmented by a detection service that measures loss, duplication and accuracy, achieving sub‑1% SDK loss, integrity tagging, comprehensive monitoring dashboards, and includes a senior data‑engineer hiring call.
Tenon is a lightweight MVVM framework built on Didi's Hummer runtime that bridges the Web ecosystem (Vue 3 and React) with native Android and iOS, offering multi‑framework support, easy adoption, extensible plugins, high availability, and a three‑layer architecture illustrated by real‑world Didi case studies.
DiDi’s real‑time MySQL‑to‑Hive pipeline captures row‑mode binlog with a custom Canal component, converts it to JSON, streams it via Kafka to HDFS, restores it into Hive tables, and uses Dquality for integrity, achieving millisecond latency for over 19,000 daily sync tasks handling roughly 50 TB of data.
