{"id":20175,"date":"2025-09-25T09:34:54","date_gmt":"2025-09-25T09:34:54","guid":{"rendered":"https:\/\/www.vdocipher.com\/blog\/?p=20175"},"modified":"2025-10-09T10:44:39","modified_gmt":"2025-10-09T10:44:39","slug":"netflix-tech-stack-and-architecture","status":"publish","type":"post","link":"https:\/\/www-uat.vdocipher.com\/blog\/netflix-tech-stack-and-architecture\/","title":{"rendered":"Netflix Tech Stack Explained: CDN (Open Connect), and Microservices"},"content":{"rendered":"

This is a source-driven explainer of the Netflix tech stack<\/strong> and Netflix architecture<\/strong>. It summarizes what Netflix has publicly shared across client surfaces (React, Falcor or GraphQL), the edge (Zuul), the Netflix microservices architecture<\/strong> on AWS (Titus, Spinnaker), data and streaming platforms (EVCache, Cassandra, Mantis, Keystone), video pipeline and codecs (per-title and shot-based with VMAF, AV1), DRM with HTML5 EME, and the Netflix CDN<\/strong>, Open Connect<\/strong>. Every claim references an official source, and anything not disclosed is labeled \u201cNot publicly documented,\u201d with compact Netflix architecture diagram<\/strong> callouts for orientation.<\/p>\n

TL;DR<\/strong><\/h2>\n

Cloud and runtime<\/strong> – Netflix completed its AWS migration in 2016 and runs thousands of services in containers on Titus<\/strong>, with global CD through Spinnaker<\/strong>.<\/p>\n

Clients<\/strong> – Web uses React<\/strong> with SSR optimizations, Falcor powered data fetching historically, GraphQL<\/strong> with the DGS<\/strong> framework is used broadly today. Consumer mobile UI framework is not comprehensively published, Studio apps share logic via Kotlin Multiplatform<\/strong>.<\/p>\n

Video<\/strong> – Per-title then shot-based encoding guided by VMAF<\/strong>, AV1<\/strong> on supported TVs, HTML5 EME on the web, delivery over Netflix Open Connect.<\/strong><\/p>\n

Edge and identity<\/strong> – Zuul<\/strong> and Zuul 2<\/strong> sit in front of device APIs, with edge authentication and token-agnostic identity propagation.<\/p>\n

Not public<\/strong> – Precise internal schemas, device-by-device DRM vendor mapping, cost models, and some operational playbooks.<\/p>\n

\n
\n

Table of Contents:<\/h2>\n
\n
    \n
  1. 1) Client Surfaces and Data Fetching (Netflix Tech Stack, client layer)<\/a><\/li>\n
  2. 2) Edge and API Gateway (Netflix architecture at the edge)<\/a><\/li>\n
  3. 3) Core Services and Orchestration (Netflix microservices architecture)<\/a><\/li>\n
  4. 4) Data Layer and Caching in the Netflix Tech Stack<\/a><\/li>\n
  5. 5) Video Pipeline, Encoding, DRM, and Delivery (Netflix CDN and Open Connect)<\/a><\/li>\n
  6. 6) Reliability, Observability, and Testing<\/a><\/li>\n
  7. 7) How to Build a Netflix-Grade Streaming Architecture in 2025 (from microservices at Netflix to your stack)<\/a><\/li>\n
  8. FAQs<\/a><\/li>\n<\/ol>\n<\/div>\n<\/div>\n<\/div>\n

    1) Client Surfaces and Data Fetching (Netflix Tech Stack, client layer)<\/h2>\n

    Web<\/strong> – Netflix adopted React for netflix.com and documented performance wins from server and client rendering optimizations.<\/p>\n

    Mobile and TV<\/strong> – Netflix has discussed Falcor on Android historically and later GraphQL for services via the DGS framework. Consumer app UI framework details are not comprehensively published. Studio mobile apps share business logic with Kotlin Multiplatform.<\/p>\n

    Falcor vs GraphQL at Netflix<\/strong> – Falcor powered multiple member UIs historically. GraphQL adoption is public via Netflix\u2019s DGS framework for building GraphQL services on Spring Boot. Studio and platform teams reference GraphQL with DGS, while Falcor remains documented in legacy posts.<\/p>\n\n\n\n\n\n\n\n\n\n
    Device surface<\/th>\nPublicly documented approach<\/th>\nSource<\/th>\n<\/tr>\n
    Web (member UI)<\/td>\nReact with server and client rendering optimizations<\/td>\ntechblog.netflix.com<\/a><\/td>\n<\/tr>\n
    Android (consumer)<\/td>\nHistorical Falcor usage noted, current consumer UI framework not comprehensively published<\/td>\ntechblog.netflix.com<\/a><\/td>\n<\/tr>\n
    iOS (consumer)<\/td>\nNot publicly documented<\/td>\nNot publicly documented<\/td>\n<\/tr>\n
    TV devices<\/td>\nReact use documented at Netflix, custom renderers referenced historically, specifics not centralized<\/td>\nnetflixtechblog<\/a><\/td>\n<\/tr>\n
    Studio mobile (Android + iOS)<\/td>\nKotlin Multi-platform for shared business logic<\/td>\n<\/td>\n<\/tr>\n
    Data fetching<\/td>\nFalcor historically, GraphQL services using DGS today<\/td>\nnetflixtechblog<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
    \n
    \n

    Explore More \u2705<\/p>\n

    VdoCipher ensures Secure Video Hosting for OTT Platforms<\/a><\/p>\n<\/div>\n

    \n

    VdoCipher helps over 3000+ customers from over 120+ countries to host their OTT videos securely, helping them to boost their video revenues.<\/p>\n<\/div>\n<\/div>\n

    2) Edge and API Gateway (Netflix architecture at the edge)<\/h2>\n

    Zuul at the edge<\/strong> – Zuul fronts device APIs. Zuul 2 re-architected the gateway on Netty for asynchronous, non-blocking I\/O and persistent connections at high scale.<\/p>\n

    Identity at the edge<\/strong> – Netflix documented edge authentication and token-agnostic identity propagation, decoupling device tokens from internal identity formats while maintaining cryptographic guarantees.<\/p>\n\n\n\n\n\n\n
    Edge component<\/th>\nRole<\/th>\nSource<\/th>\n<\/tr>\n
    Zuul<\/td>\nAPI gateway in front of device APIs, routing and protection<\/td>\ntechblog.netflix.com<\/a><\/td>\n<\/tr>\n
    Zuul 2<\/td>\nNetty-based async gateway for high throughput and long-lived connections<\/td>\ntechblog.netflix.com<\/a><\/td>\n<\/tr>\n
    Edge auth<\/td>\nToken-agnostic identity propagation from the edge into services<\/td>\nnetflixtechblog.com<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    3) Core Services and Orchestration (Netflix microservices architecture)<\/h2>\n

    Containers and runtime<\/strong> – Netflix runs containers on Titus, its internal container platform open sourced in 2018.<\/p>\n

    Service discovery and resilience<\/strong> – Netflix publicly documented Eureka for service discovery, and Hystrix for latency and fault tolerance.<\/p>\n

    Global CD<\/strong> – Spinnaker enables multi-region continuous delivery, introduced and documented by Netflix.<\/p>\n

    Workflows<\/strong> – Conductor orchestrates microservice workflows.<\/p>\n\n\n\n\n\n\n\n\n
    Runtime or platform<\/th>\nWhat it does<\/th>\nSource<\/th>\n<\/tr>\n
    Titus<\/td>\nContainer management and execution platform at Netflix<\/td>\nnetflixtechblog.com<\/a><\/td>\n<\/tr>\n
    Eureka<\/td>\nREST-based service discovery in AWS<\/td>\ntechblog.netflix.com<\/a><\/td>\n<\/tr>\n
    Hystrix<\/td>\nCircuit breaking and latency tolerance (historical usage)<\/td>\ntechblog.netflix.com<\/a><\/td>\n<\/tr>\n
    Spinnaker<\/td>\nGlobal continuous delivery<\/td>\ntechblog.netflix.com<\/a><\/td>\n<\/tr>\n
    Conductor<\/td>\nMicroservice workflow orchestration<\/td>\ntechblog.netflix.com<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    4) Data Layer and Caching in the Netflix Tech Stack<\/h2>\n

    Caches and KV access<\/strong> – EVCache, a memcached-based system, is used for ephemeral, volatile caching. Netflix has also documented a Key-Value Data Abstraction Layer for standardizing access to backing stores like Cassandra.<\/p>\n

    Datastores<\/strong> – Cassandra appears frequently in Netflix posts and talks as a primary online store for scale-out workloads.<\/p>\n

    Real-time and batch<\/strong> – Mantis is the stream processing platform for operational telemetry. Keystone is the real-time data movement and processing backbone. Genie orchestrates big-data jobs.<\/p>\n\n\n\n\n\n\n\n\n\n
    Data component<\/th>\nWorkload<\/th>\nSource<\/th>\n<\/tr>\n
    EVCache<\/td>\nLow-latency ephemeral caching across regions<\/td>\ntechblog.netflix.com<\/a><\/td>\n<\/tr>\n
    Cassandra<\/td>\nDurable online data store at large scale<\/td>\nnetflixtechblog.com<\/a><\/td>\n<\/tr>\n
    KV Abstraction<\/td>\nStandardized access layer over key-value stores<\/td>\nnetflixtechblog.com<\/a><\/td>\n<\/tr>\n
    Mantis<\/td>\nReal-time stream processing<\/td>\ntechblog.netflix.com<\/a><\/td>\n<\/tr>\n
    Keystone<\/td>\nReal-time data pipeline<\/td>\nnetflixtechblog.com<\/a><\/td>\n<\/tr>\n
    Genie<\/td>\nBig-data job orchestration<\/td>\nnetflix.github.io<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n


    \n\"Secure
    \n<\/a><\/p>\n

    5) Video Pipeline, Encoding, DRM, and Delivery (Netflix CDN and Open Connect)<\/h2>\n

    Encoding<\/strong> – Netflix pioneered per-title encoding, later shipped optimized shot-based encodes, and measures perceptual quality with VMAF. AV1 streaming is enabled on supported TVs.<\/p>\n

    Web playback and DRM<\/strong> – Netflix publicly details its migration to HTML5 playback using MSE and EME across major browsers.<\/p>\n

    Delivery via Netflix Open Connect<\/strong> – Open Connect is Netflix\u2019s own CDN, with Open Connect Appliances embedded at ISPs and at interconnect sites.<\/p>\n\n\n\n\n\n\n\n\n
    Stage<\/th>\nPublicly stated Netflix approach<\/th>\nSource<\/th>\n<\/tr>\n
    Ingest<\/td>\nNot publicly documented<\/td>\nNot publicly documented<\/td>\n<\/tr>\n
    Transcode<\/td>\nPer-title, then per-shot optimization guided by perceptual metrics (VMAF)<\/td>\ntechblog.netflix.com<\/a><\/td>\n<\/tr>\n
    Package<\/td>\nNot publicly documented<\/td>\nNot publicly documented<\/td>\n<\/tr>\n
    Protect<\/td>\nHTML5 playback via EME on web, platform DRMs per device<\/td>\nnetflixtechblog.com<\/a><\/td>\n<\/tr>\n
    Deliver<\/td>\nOpen Connect CDN program and overview documentation<\/td>\nopenconnect.netflix.com<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    6) Reliability, Observability, and Testing<\/h2>\n

    Chaos engineering<\/strong> – Netflix created the Simian Army, extended with FIT for precise failure injection and ChAP for automated, guarded chaos experiments.<\/p>\n

    Telemetry<\/strong> – Atlas is Netflix\u2019s primary dimensional time-series telemetry system, with streaming evaluation for alerting.<\/p>\n

    Adaptive concurrency<\/strong> – Netflix open-sourced concurrency-limits for adaptive concurrency control, a practical load-shedding primitive.<\/p>\n

    Canary analysis<\/strong> – Kayenta provides automated canary analysis in the deployment workflow.<\/p>\n\n\n\n\n\n\n\n\n\n
    Practice or tool<\/th>\nWhat it solves<\/th>\nSource<\/th>\n<\/tr>\n
    Simian Army<\/td>\nFault tolerance testing through controlled failure<\/td>\ntechblog.netflix.com<\/a><\/td>\n<\/tr>\n
    FIT<\/td>\nPrecise failure injection for services and requests<\/td>\nnetflixtechblog.com<\/a><\/td>\n<\/tr>\n
    ChAP<\/td>\nAutomated chaos experiments with guardrails<\/td>\nnetflixtechblog.com<\/a><\/td>\n<\/tr>\n
    Atlas<\/td>\nHigh-scale metrics and near real-time ops insight<\/td>\ntechblog.netflix.com<\/a><\/td>\n<\/tr>\n
    concurrency-limits<\/td>\nAdaptive concurrency and load shedding<\/td>\ngithub.com<\/a><\/td>\n<\/tr>\n
    Kayenta<\/td>\nAutomated canary analysis<\/td>\nnetflixtechblog.com<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    7) How to Build a Netflix-Grade Streaming Architecture in 2025 (from microservices at Netflix to your stack)<\/h2>\n

    This section turns Netflix\u2019s public architecture into a buildable blueprint. It shows the effort required if you assemble the stack yourself, then also shows development architecture with ready to use SaaS like VdoCipher<\/a>.<\/p>\n

    7.1 Custom build streaming app tech stack (Heavy Tech Load)<\/h3>\n

    \"Streaming<\/a>You own and operate every moving part end-to-end. Typical build list:<\/p>\n