Uber tackled the challenge of replicating over 350 PB of data across on‑premise and cloud lakes by redesigning Hadoop Distcp, moving intensive tasks to the Application Master, parallelising copy‑listing and commit phases, and leveraging Uber‑mapper jobs to dramatically cut latency and improve resource efficiency.
This article explains how Uber built a robust, continuously backed‑up and recoverable database platform that handles tens of petabytes of data, detailing the challenges, architecture, scheduling, backup and restore frameworks, and the technology‑specific snapshot logic that enable fast RPO/RTO at massive scale.
This article explains how Uber solves the "nearby driver" problem by employing the open‑source H3 hexagonal spatial index, hierarchical grids, Cassandra for persistent storage, and Redis caching to deliver fast, accurate, and scalable real‑time location services.
This whitepaper analyzes the shift of data‑intensive analytics to cloud‑native platforms, examines Uber Presto’s fragmented I/O patterns, reveals hidden storage‑API cost impacts, and proposes cloud‑aware I/O optimization strategies to improve performance‑cost efficiency.
Uber reduced cluster-wide CPU allocation by 11% and stabilized P99 latency by replacing CPU quota throttling with cpusets, a CPU‑pinning technique that isolates containers on specific cores, eliminating throttling while only slightly increasing P50 latency.
After more than a year of effort, Uber migrated its entire experimentation and feature‑flag ecosystem—including thousands of developers, dozens of partners, multiple mobile apps, and hundreds of services—to a new, unified platform that improves reliability, flexibility, and data quality while retiring over 50,000 legacy experiments.
The article details Uber's transition from a monolithic codebase to a service‑oriented architecture, describing its goals, functional and non‑functional requirements, DISCO dispatch optimization, micro‑service communication, data pipelines, storage choices, and supporting components such as Kafka, Hadoop, Spark, and security mechanisms.
Uber transitioned from a monolithic codebase to a service‑oriented architecture, detailing goals like 99.99% reliability, core vs optional code separation, DISCO dispatch optimization, real‑time location tracking with Kafka, Hadoop, Spark, and the supporting infrastructure such as load balancers, web sockets, and security layers.
Uber’s self‑service API gateway provides a unified entry point for all applications, handling routing, protocol translation, rate limiting, authentication, and more, with configurable components such as protocol managers, middleware, handlers, and clients defined via YAML/Thrift, and leverages code generation and a DAG‑based build system to produce scalable, extensible backend services.
The article provides an in‑depth overview of Uber’s self‑service API gateway, describing its role in microservice architectures, the configurable components such as protocol manager, middleware, handlers and clients, the YAML/Thrift‑based configuration and code‑generation pipeline, and the advanced features and challenges involved in building a production‑grade gateway.
This article explains Uber’s Domain‑Oriented Microservice Architecture (DOMA), describing why the company adopted it, how domains, layer design, gateways and extensions are organized, the operational benefits achieved, and practical advice for companies of different sizes considering a similar evolution.
This article explains Uber's Domain‑Oriented Microservice Architecture (DOMA), describing why microservices were adopted, the challenges of scaling them, the concepts of domains, layers, gateways and extensions, and provides practical advice for organizations looking to reduce complexity while retaining flexibility.
Uber’s Domain‑Oriented Microservice Architecture (DOMA) tackles the growing complexity of large‑scale microservice systems by organizing services into domains, layers, and gateways, offering improved reliability, clearer ownership, and faster development, while providing practical guidance for teams adopting similar architectures.
The article describes Uber’s Gairos platform—a unified real‑time data ingestion, storage, and query system built on Kafka, Elasticsearch and RT‑Gairos—detailing its architecture, Uber use cases such as dynamic pricing, the scalability and reliability challenges faced, and the suite of optimization techniques (sharding, query routing, caching, index merging, template tuning, and data‑pruning) applied to achieve low latency and high throughput.
This article explains Uber's Domain‑Oriented Microservice Architecture (DOMA), describing why microservices were adopted, how DOMA structures domains, layers, gateways and extensions, the benefits observed, and practical advice for companies of different sizes considering a similar evolution.
This article explains the design of Apache Hudi, its core concepts such as upserts and incremental pulls, how Uber leverages it for large‑scale data‑lake operations, the enhancements made over time, and the project’s future plans within the Apache ecosystem.
Uber open‑sourced Piranha, an automated tool that scans codebases, generates refactoring diffs, and removes obsolete feature‑flag code, helping teams reduce technical debt, improve performance, and accelerate development cycles across mobile platforms.
The article argues that simple, term‑free design processes—focused on business problems, whiteboard brainstorming, concise documentation, and iterative feedback—outperform formal architecture tools, patterns, and dedicated architects in large‑scale systems like Uber, emphasizing engineering culture over rigid methodologies.
This article examines how Uber built and scaled its real-time ride-sharing platform, detailing the original simple PHP-MySQL architecture, subsequent extensions with message queues, MongoDB, Ringpop storage, TChannel communication, fault-tolerance strategies, latency challenges, and practical tools for distributed system design.
Uber’s engineering team explains how Postgres’s immutable rows, write‑ahead logging, and replication design caused write amplification, upgrade pain, and data‑corruption risks, leading them to adopt a MySQL‑based Schemaless sharding layer that offers better buffering, connection handling, and replication semantics for massive scale.
This article explains the core concepts of microservice architecture, outlines its components and pros‑and‑cons, and uses Uber's evolution from a monolithic system to a distributed microservice platform as a detailed case study.
Microsoft has launched an open‑source tool that lets distribution maintainers and developers create custom Linux packages for the Windows Subsystem for Linux, while also previewing new WSL features such as background tasks and Unix‑domain sockets, and reporting that Uber disabled safety radar on a Volvo autonomous vehicle involved in a fatal pedestrian crash.
A fully autonomous Uber‑operated Volvo XC90 struck and killed pedestrian Elaine Herzberg in Tempe, marking the first pedestrian fatality involving a self‑driving car and prompting investigations, regulatory scrutiny, and a temporary halt to Uber’s and other companies’ public‑road autonomous testing.
This article examines Uber's rapid 38‑fold growth by detailing the design, scaling techniques, and fault‑tolerance mechanisms of its real‑time market platform, including geographic indexing, microservices, distributed storage, and the DISCO scheduling system.
Uber built Schemaless, a MySQL‑backed, sharded JSON datastore with immutable cells, triggers, and secondary indexes, to overcome PostgreSQL limits and achieve linear scalability, high write throughput, reliable change notifications, and operational resilience for its ride‑hailing platform.
Dynamic pricing strategy, as used by Uber and other sharing‑economy platforms, adjusts prices to balance supply and demand, leveraging economic incentives, algorithmic calculations, and real‑time data, and its benefits, drawbacks, and practical applications across ride‑hailing, delivery, and crowdsourcing services are examined.
The article critically examines Uber's claim that MySQL outperforms PostgreSQL for heavy‑write workloads, explaining the nuances of update costs, HOT tuples, index handling, replication bugs, and why the original piece should not be taken as generic database advice.
This article breaks down Uber’s hybrid‑cloud infrastructure, storage choices, logging pipeline, service discovery, development languages, deployment tools, and monitoring system, revealing how the company builds a highly available, low‑latency platform that powers its global ride‑hailing service.
Uber’s adoption of microservices showcases both the flexibility of using multiple languages and independent release cycles, while also exposing challenges such as duplicated effort across teams, type‑unsafe JSON interfaces, and the need for rigorous failure testing, offering valuable lessons for large‑scale system design.
This article details Uber's transition to a modular microservice architecture, the RFC process for new services, and the design, implementation, and operational lessons learned from creating Tincup, a high‑performance currency and exchange‑rate service supporting dozens of currencies worldwide.
This article details Uber's transition to a modular microservice architecture, describing the RFC process for new services, the design and implementation of the Tincup currency service, and the technologies—such as MVCS, UDR, Tornado, TChannel, and Thrift—used to ensure scalability, reliability, and efficient production deployment.
Uber’s Schemaless datastore, described in the Mezzanine project, replaces a single PostgreSQL node with a highly available, horizontally scalable MySQL‑backed key‑value store, detailing its architecture, data model, triggers, indexing, fault‑tolerance, and operational considerations for massive trip data.
This article explains how Uber engineered Schemaless, a highly available, horizontally scalable datastore built on MySQL that stores immutable JSON cells without a fixed schema, detailing its design goals, architecture, data model, trigger system, indexing strategy, and fault‑tolerant read/write mechanisms.
Uber’s Schemaless is a highly available, horizontally scalable MySQL‑based key‑value store that abandons fixed schemas, supports JSON blobs, triggers, and global secondary indexes, and was created to meet five critical requirements for trip data storage, including linear expansion, write throughput, and reliable operations.
This article examines Uber's rapidly growing real‑time dispatch platform, detailing its geo‑spatial indexing, microservice architecture, fault‑tolerant design, and scaling techniques that enable millions of writes per second and high availability across thousands of nodes.
Uber’s CTO Thuan Pham revealed at a Chinese tech salon how the company’s global architecture, data‑center strategy, cloud partnership with Baidu, anti‑fraud machine‑learning models, map localization and big‑data analytics together enable a unified yet locally adapted ride‑sharing platform across China and the world.
Uber’s CTO Thuan Pham explains how the company builds a specialized engineering team and core infrastructure in China, uses data‑center support, dynamic pricing, anti‑fraud machine learning, and safety features to adapt its global platform to local market needs.
The article argues that Uber’s consumer‑centric, supply‑demand integration and its new Operator shopping service illustrate why a true company focused on creating value for customers outperforms platform‑only giants like Alibaba, which risk becoming mere market managers.
This article examines Uber's rapid 38‑fold growth and the engineering choices behind its real‑time market platform, detailing the scheduling system, geographic indexing, microservices, Ringpop, TChannel, and strategies for scalability, availability, and fault tolerance.
This article details Uber's rapid growth and the engineering choices behind its real‑time dispatch platform, covering geospatial indexing, microservice architecture, scaling techniques like Ringpop and TChannel, and strategies for availability and fault tolerance.
