Rapid Event Notification System (RENO) at Netflix: Design, Architecture, and Lessons Learned

Netflix serves over 220 million active members who interact with the service across many devices and platforms. To ensure a seamless experience, Netflix created the Rapid Event Notification (RENO) system, a scalable, extensible service that enables server‑initiated communication with devices for a variety of use cases.

Motivation

Rapid membership growth and increasing system complexity pushed Netflix toward an asynchronous architecture that supports both online and offline computation. Traditional request‑response patterns could not satisfy the need to instantly inform devices about user‑driven changes or experience updates across iOS, Android, smart TVs, Roku, Fire Stick, browsers, and more.

User Scenarios

Watch activity – updating "continue watching" lists on all devices.

Personalized experience refresh – delivering timely recommendation updates.

Membership plan changes – reflecting plan modifications instantly.

My List updates – synchronizing additions or removals across devices.

Profile changes – propagating account‑setting updates.

System diagnostic signals – sending troubleshooting data to apps.

Design Decisions

Single event source – using Netflix's internal Manhattan distributed computation framework as an indirect layer to aggregate events from multiple micro‑services.

Event priority determination – assigning priorities to use cases and routing them to priority‑specific queues and processing clusters.

Hybrid communication model – combining push (immediate server‑initiated notifications) and pull (device‑initiated callbacks) to accommodate always‑online mobile devices and intermittently‑online TV devices.

Targeted delivery – filtering notifications per device type to limit traffic.

High RPS management – employing age‑based filtering, online‑device tracking via Zuul, aggressive auto‑scaling, deduplication, and batched sending to downstream push services (APNS, FCM, etc.).

Single Event Source

Netflix leverages Manhattan’s event management framework to create an indirect layer that serves as RENO’s unified event source, consolidating streams from various internal systems.

Event Priority Determination

Events are categorized by source and importance; higher‑priority events (e.g., profile maturity changes) are processed before lower‑priority ones (e.g., diagnostic signals) by routing them to dedicated queues and clusters.

Hybrid Communication Model

Because mobile devices are usually online while smart TVs are often offline, RENO uses a mixed push‑and‑pull approach: servers push notifications when possible, and devices pull updates during appropriate lifecycle phases, ensuring coverage for devices that cannot receive pushes.

Targeted Delivery

RENO evaluates the target device type for each actionable event and delivers notifications only to eligible devices, dramatically reducing unnecessary traffic.

Managing High RPS

During peak periods RENO handles ~150 k events per second. Optimizations include age‑based filtering to discard stale events, sending only to currently online devices, aggressive scaling policies, deduplication of duplicate events, and parallel batched sending to external push services.

Architecture Diagram

As shown below, RENO consists of several components:

Event Triggers

Member actions and system‑driven updates that require a refreshed experience on devices.

Event Management Engine

Manhattan listens for specific events and forwards them to appropriate queues.

Priority‑Based Queues

AWS SQS queues populated according to priority rules defined in Manhattan.

Priority‑Based Clusters

AWS instance clusters subscribe to queues of matching priority, process events, and generate actionable notifications.

Outbound Messaging System

Netflix’s messaging layer (including a custom Zuul Push solution) delivers notifications to mobile, TV, and other streaming devices.

Persistent Storage

Cassandra stores all notifications per device, allowing devices to poll at their own pace.

Observability

Comprehensive metrics, alerts, and logs (including service‑edge metrics) are collected; Mantis processes real‑time streams to surface device‑level anomalies, and platform‑specific alerts aid rapid root‑cause analysis.

Benefits

Easy support for new use cases.

Horizontal scalability to higher throughput.

Initially focused on personalized experience updates, RENO quickly evolved into a centralized rapid‑notification platform for all Netflix product areas, enabling plug‑and‑play addition of new use cases and supporting mixed delivery across platforms.

Future Outlook

As membership continues to grow, RENO will further enhance the Netflix experience by exploring guaranteed delivery, batch processing, and additional features that reduce communication footprints while opening new use cases.

References

[1] Original article: https://netflixtechblog.com/rapid-event-notification-system-at-netflix-6deb1d2b57d1