Twitter Read Path System Architecture

Figure 1: Original whiteboard schematic illustrating the "Read Path" routing from client applications through API layers, Timeline/Home Mixers, down to ranking, scoring, and storage engines.

Twitter Read Path Architecture Diagram

graph TD
                            Client[Clients: Web App, Android, iPhone] --> FE[Twitter Front End]
                            FE --> GraphQL[GraphQL / Federated Router]
                            FE --> TLSAPI[HTTP TLS-API - deprecated]

                            GraphQL --> Strato[Strato Column]
                            GraphQL --> TweetHydration[Tweet/User Content Hydration]
                            GraphQL --> Visibility[Visibility Filtering]

                            TLSAPI --> TimelineMixer
                            GraphQL --> TimelineMixer

                            TimelineMixer --> HomeMixer[Home Mixer]
                            TimelineMixer --> Ads[Ad Mixer]
                            TimelineMixer --> People[People Discovery]

                            HomeMixer --> Manhattan[(Manhattan KV Store)]
                            HomeMixer --> Gizmoduck[Gizmoduck User Profile]
                            HomeMixer --> SocialGraph[Social Graph]
                            HomeMixer --> TweetPie[Tweety Pie]
                            HomeMixer --> HomeRanker[Home Ranker]

                            HomeRanker --> HomeScorer[Home Scorer]
                            HomeScorer --> Prediction[Prediction Service]

                            HomeRanker --> CrMixer[CrMixer]
                            HomeRanker --> EarlyBird[EarlyBird Search]
                            HomeRanker --> Uteg[Uteg Engagement Graph]
                            HomeRanker --> Space[Space Audio]

                            classDef client fill:#f9f,stroke:#333,stroke-width:2px;
                            classDef frontend fill:#ff9,stroke:#333,stroke-width:2px;
                            classDef router fill:#d4f1f4,stroke:#333,stroke-width:2px;
                            classDef mixer fill:#befabe,stroke:#333,stroke-width:2px;
                            classDef ranker fill:#ffcc99,stroke:#333,stroke-width:2px;
                            classDef storage fill:#ff9999,stroke:#333,stroke-width:2px;
                            classDef service fill:#e1bee7,stroke:#333,stroke-width:1px;

                            class Client client;
                            class FE frontend;
                            class GraphQL,TLSAPI router;
                            class TimelineMixer,HomeMixer,Ads,People mixer;
                            class HomeRanker,HomeScorer ranker;
                            class Manhattan storage;
                            class Strato,TweetHydration,Visibility,Gizmoduck,SocialGraph,TweetPie,Prediction,CrMixer,EarlyBird,Uteg,Space service;

Figure 2: A color-coded and corrected diagram reconstructing the data flow and system hierarchy, properly representing Home Ranker within the stack.

Next-Gen System Direction

The architecture outlines a transition path (marked by dashed lines) moving towards a consolidated structure involving unified mixers, Thrift RPC-based downstream scoring, and the deprecation of legacy TLS-API HTTP pathways in favor of federated GraphQL layers.

1. Clients & Edge API

Supported Platforms

Web
Android
iPhone

Gateway Entry

Twitter Front-end

Protocols / Router Pathways

GraphQL / Federated (iPhone, Web)
- Strato Column
- Tweet / User Content Hydration
- Visibility Filtering
HTTP TLS-API (Android)
Being Deprecated

2. Timeline Mixer

Responsible for orchestrating timeline assembly and applying initial product rules.

Key Functions

Inject Ads, Who-to-Follow, Onboarding
Conversation Module
Cursoring / Pagination
Tweet Deduplication
Served Data Logging

Upstream Services

People Discovery Service
Ad Mixer
Onboarding Service

3. Home Mixer

Handles home timeline generation and routes directly to key key-value stores and core entities.

Direct Downstream Stores

Manhattan (Distributed Key-Value Store)
Gizmoduck (User Profile Service)
Social Graph (User connections & graph data)
Tweety Pie (Tweet metadata & hydration)

4. Home Ranker & Scorer

Coordinates candidate generation, filtering, and machine learning scoring via Thrift RPC.

Home Ranker

Candidate Fetch

Home Scorer

Prediction Service
Feature Hydration
- Manhattan
- Memcache

5. Timeline & Scorer Services

Timeline Scorer

Dedicated system for evaluating, ranking, and ordering timeline items based on heavy scoring pipelines.

Timeline Service

Manages stateful timeline representations and caches for quick retrieval.

6. Candidate Services

Pipeline Stages

Fetch → Feature → Scoring

Source Service Engines

CrMixer
EarlyBird (Search Index / Tweet Retrieval)
Uteg (User Tweet Engagement Graph)
Space (Audio Spaces candidate retrieval)
Community

Deep Downstream

Prediction Service