How Medium Scaled with Microservices: Principles and Real-World Strategies

Microservice architecture aims to help technical teams deliver products faster, safer, and with higher quality; decoupling services enables rapid iteration with minimal system impact.

Medium, founded by a Twitter co‑founder in 2012 as a large social blogging platform, started as a single Node.js monolith. As the system grew and the team expanded, it began migrating to a microservice architecture in early 2018.

Microservice Principles

In a microservice architecture, many loosely coupled services work together, each focusing on a single purpose with highly cohesive data and behavior.

Three important design principles:

Single Purpose – Each service concentrates on one goal and does it well.

Loose Coupling – Services know little about each other; changes in one service do not require changes in others, and communication occurs only via service interfaces.

High Cohesion – Services encapsulate related behavior and data; new features should involve modifications to only one service.

Following these three guidelines enables microservices to realize their full potential. Ignoring them leads to higher costs, complex deployments, and a distributed monolith.

Medium's Microservice Strategy

(1) Build New Services with Clear Value

Never create a service without clear product or engineering value. A new service must deliver tangible benefits to users or improve the development workflow; otherwise, continue evolving the monolith.

(2) Avoid Shared Persistent Storage

Sharing databases across services exposes implementation details, violating loose coupling and high cohesion. Each service should own its data, exposing it via APIs or read‑only replicas.

In a microservice architecture, a specific type of data should be owned by a single service; other services access it through APIs or maintain read‑only copies.

Example: a recommendation service needing post data.

Monolithic approach – direct access to the post table – suffers from cache invalidation, database lock‑in, duplicated business logic, and tight coupling.

Decoupled alternatives:

Solution A – a dedicated post service owns the data; other services call its API.

Solution B – the monolith publishes changes via a message queue to the recommendation service.

Solution C – an ETL pipeline creates a read‑only replica of post data for the recommendation service.

All three allow the recommendation service to own its data, enabling flexible caching and database choices.

(3) Services Focus on Their Own Business Logic

Each service should handle only its responsibilities, leaving networking, protocol, deployment, and monitoring to dedicated infrastructure.

Network – Service discovery, routing, load balancing are handled by a Service Mesh (Medium uses Istio and Envoy).

Communication Protocol – Medium chose gRPC for efficient, cross‑platform RPC, avoiding the overhead of HTTP REST+JSON for server‑to‑server calls.

Deployment – All services run in containers orchestrated by Kubernetes, with a unified build‑test‑package‑deploy pipeline.

(4) Consistent and Comprehensive Monitoring

Monitoring (logs, metrics, dashboards, alerts) is essential for understanding system behavior and troubleshooting.

Medium initially faced two problems: lack of observability due to microservice complexity, and fragmented, duplicated monitoring efforts across teams.

Complexity made monitoring difficult, reducing system visibility.

Team‑specific monitoring tools caused fragmentation and low observability.

Medium adopted DataDog for automated dashboards, alerts, and logs, and LightStep for performance tracing.

(5) Embrace Failure

Medium emphasizes failure testing, robust error handling, and automated recovery mechanisms.

Maintain awareness of potential errors.

Invest in RPC error handling.

Make failures observable.

Perform failure testing before releasing new services and maintain checklists.

Build automatic recovery mechanisms.

This article is translated and compiled from the original Medium Engineering post: Microservice Architecture at Medium .