What Uber’s Microservices Reveal About the Pros and Cons of Distributed Architecture

Pros and Cons of the Microservice Model

Microservices allow teams to use different programming languages; Uber, for example, runs services in Node.js, Python, Go, Java, and Scala.

This approach lets each team control its own release cycle and be responsible for its services online.

However, because each team works independently, overall development speed can suffer—for instance, a Java team must define communication protocols that the Node.js and Go teams must also implement.

Similarly, bugs fixed on one platform often need to be addressed again on other platforms, leading to duplicated effort.

Insufficient estimation of the cost of mixing multiple languages can exceed expectations.

For new hires, microservices reduce system‑understanding difficulty, enabling rapid onboarding, as demonstrated by Uber’s rapid growth.

Uber’s Overall Architecture

Uber operates many data centers worldwide to bring users closer to data and improve availability.

The dispatch system, initially built in Node.js, scaled to such a size that it was gradually migrated to Go and Java.

High‑compute services, such as mapping, are primarily written in Java to handle intensive calculations for ETA and route planning.

Task state management in the dispatch system relies on a Riak cluster, an Erlang‑based highly scalable distributed datastore.

Completed tasks are removed from the dispatch system and streamed to other business systems via Kafka.

All Kafka streams feed into Hadoop for data analysis.

Quality Assurance

Uber places great emphasis on system quality; failure testing is a critical activity, and all systems are designed to be verifiable.

The failure testing framework is based on Netflix’s Simian Army, customized for Uber’s needs.

Simian Army performs random fault injection to validate reliability.

Establishing a simulated environment and a comprehensive integration test suite are essential parts of the verification process.

Through verification, Uber discovered that many interfaces used untyped JSON, which is not type‑safe and makes validation difficult, increasing verification costs.

Moving forward, Uber plans to refactor interfaces to use type‑safe, verifiable formats.

In addition to extensive internal testing, Uber maintains a large global mobile testing team that conducts black‑box testing from a user perspective.

Source: Interview with Uber chief architect Matt Ranney at QCon 2016.