Design Uber's Real-Time Dispatch & Schemaless MySQL Storage

1. Problem Statement & Scope Clarification

System Mission

Design Uber's real-time geospatial dispatch engine and distributed trip data persistence layer. The platform must process millions of continuous driver GPS telemetry pings per second, compute sub-millisecond supply-demand matching using hexagonal geospatial partitioning (Uber H3), coordinate the end-to-end trip lifecycle state machine, and store immutable trip records across a custom sharded append-only storage engine on MySQL (Schemaless).

Functional Requirements

1. High-Frequency Telemetry Ingestion (DriverLocationPing): Ingest GPS coordinates from 6M+ drivers every $4\text{ seconds}$ with $P99 < 10\text{ ms}$.
2. Geospatial Supply/Demand Matching (DispatchRider): Given a rider's pickup latitude/longitude, locate and rank the top $K=10$ nearest available drivers in $< 20\text{ ms}$ using H3 hexagonal spatial indexing.
3. Trip Lifecycle State Machine: Coordinate state transitions (REQUESTED $\rightarrow$ MATCHED $\rightarrow$ ARRIVED $\rightarrow$ IN_PROGRESS $\rightarrow$ COMPLETED) with strict idempotency and zero lost transitions.
4. Append-Only Immutable Trip Storage (Schemaless): Persist structured JSON trip payloads without schema migrations, supporting cross-shard secondary indexing.

Non-Functional Requirements (SLAs & SLOs)

• High Availability: $99.999\%$ uptime for dispatching and location tracking.
• Low Match Latency: $P50 < 10\text{ ms}$, $P99 < 35\text{ ms}$ for dispatch ring execution.
• Zero Loss of Financial Trip Data: Absolute durability across Multi-AZ storage cells.

2. Capacity & Scale Estimation (Back-of-the-Envelope Math)

Telemetry Scale

• Active Drivers Worldwide: $6,000,000$.
• Location Ping Frequency: Every $4\text{ seconds}$.
• Peak Telemetry Ingestion Throughput: $\text{Peak Ingest QPS} = \frac{6,000,000\text{ drivers}}{4\text{ s}} = \mathbf{1,500,000\text{ QPS}}$
• Location Payload Size: $\approx 128\text{ bytes}$ (driver_id, lat, lng, bearing, speed, h3_index, timestamp).
• Telemetry Ingestion Bandwidth: $\text{Ingest Bandwidth} = 1,500,000 \times 128\text{ bytes} = 192,000,000\text{ B/s} \approx \mathbf{1.536\text{ Gbps}}$

Trip Persistence Scale

• Daily Completed Trips: $30,000,000\text{ trips/day}$.
• Average Trip Record Size (Schemaless Payload): $8\text{ KB}$ (includes route coordinates, pricing breakdowns, audit timestamps).
• Daily Storage Growth: $\text{Daily Storage} = 30,000,000 \times 8\text{ KB} = \mathbf{240\text{ GB/day}}$
• 5-Year Retention Storage (with 3x replication): $\text{5-Year Storage} = 240\text{ GB} \times 365 \times 5 \times 3 \approx \mathbf{1.314\text{ PB NVMe SSD}}$

3. High-Level Architecture & Component Mapping

Interactive Architecture Diagram
100%Fit100%
Synthesizing vector architecture diagram...