Project

Voila Cabs platform scaling

99.9%+ uptime and thousands of concurrent users after replacing single-instance databases with clustered MySQL and sharded MongoDB on AWS.

  • Tech Integrity Services
  • 2019
  • Ride-hailing platform
  • voilacabs.com
99.9%+Uptime
1000sConcurrent users
3×DB replicas
5×Scale headroom

Context

Product
Ride-hailing platform — iOS/Android apps, web booking, driver admin panel.
Constraint
Single-instance MySQL and MongoDB. Single point of failure; peak-hour degradation under load.
Requirement
Clustered infrastructure with automatic failover — migration on live traffic, no user-visible cutover.

Role & scope

Started on iOS; moved to backend when infrastructure became the bottleneck. Owned database clustering, AWS deployment, and failover design.

  • Designed 3-replica MySQL and MongoDB clusters with automatic primary promotion
  • Implemented MongoDB sharding for write distribution on booking and location data
  • Built AWS infrastructure for cluster hosting, monitoring, and elastic capacity
  • Coordinated with mobile and web teams for zero user-visible migration

Key decisions

Replica clusters over vertical scaling

Three-replica clusters with hot secondaries enable automatic promotion on primary failure.

Rejected: larger single instance · managed RDS only (limited for MongoDB sharding)

MongoDB sharding for write distribution

Booking and driver-location collections outgrew a single node. Sharding accepted for linear write scale.

Rejected: per-city databases · Redis-only caching (insufficient at peak)

AWS as hosting foundation

Cloud-native cluster management and monitoring without building ops tooling from scratch.

Trade-off: AWS vendor coupling vs. self-managed bare metal cost

Architecture

┌──────────┐ ┌──────────┐ ┌──────────┐ │ Mobile │ │ Web │ │ Admin │ │ iOS/And. │ │ App │ │ Panel │ └────┬─────┘ └────┬─────┘ └────┬─────┘ └─────────────┼─────────────┘ ▼ ┌────────────────┐ │ API Layer │ │ Express · PHP │ │ Socket.io │ └───────┬────────┘ │ ┌────────────┼────────────┐ ▼ ▼ ▼ ┌─────────┐ ┌─────────┐ ┌─────────┐ │ MySQL │ │ MongoDB │ │ Firebase│ │ 3-node │ │ sharded │ │ config │ │ cluster │ │ cluster │ │ │ └─────────┘ └─────────┘ └─────────┘ └────────────┬────────────┘ ▼ ┌───────────┐ │ AWS │ └───────────┘
MySQL Cluster MongoDB AWS Socket.io Express.js

Results

99.9%+Platform uptime after clustering
1000sConcurrent users without degradation
0Data loss incidents after redundant backup architecture
5–10×Headroom for marketing-driven user growth

Retrospective

  • Would align infra investment with user-growth curve earlier — SPOF risk was latent until peak load.
  • Failover testing should run on a schedule, not only after incidents.