FlashGuard

The Concurrency Challenge

When 10,000+ users hit a single "Buy" button simultaneously, traditional architectures fail in two specific ways:

Race Conditions

Two users see "1 item left" at the exact same millisecond. The database validates both, resulting in -1 inventory and a failed business promise.

Database Deadlocks

Standard SQL row-locking creates a massive queue. The CPU spikes to 100%, the connection pool exhausts, and the entire platform goes offline.

↓

FlashGuard solves this by shifting the "Truth" to a Distributed Atomic Layer.

↓

Concurrency Simulator

Environment: AWS-Lambda-Production

Total Requests

Successful Locks

Inventory Oversold

Avg Latency

0ms

[SYSTEM] Initializing Redis Lua Engine...

[SYSTEM] Connection established with Supabase PostgreSQL.

The Technical Stack

Compute Layer

Spring Boot / AWS Lambda

Handles high-concurrency business logic and scales execution based on demand.

Concurrency Layer

Redis (Lua Scripts)

Provides sub-millisecond atomic locking to prevent double-selling inventory.

Data Layer

PostgreSQL (Supabase)

Final source of truth for orders with ACID-compliant relational integrity.

Delivery Layer

Next.js / Tailwind

Real-time state updates and low-latency interaction for the end-user.

System Architecture

A distributed approach designed for high-availability. Traffic is managed in memory before being persisted to the relational database.

Client

Next.js / Tailwind

API Layer

Spring Boot / AWS

Concurrency

Redis / Lua

Persistence

PostgreSQL

Kernel Logic

Atomic Consistency via Lua

To prevent "Ghost Inventory," I moved the validation logic into a Single-Threaded Lua Script. This ensures the "Check" and "Decrement" operations are executed as one unbreakable unit within Redis.

Zero Race Conditions: No two requests can interrupt the script.
Sub-1ms Execution: Logic runs in-memory near the data.

reserve_stock.lua

-- Redis Lua Script for Atomic Reservation
local stock = tonumber(redis.call('get', KEYS[1]))

if stock > 0 then
    redis.call('decr', KEYS[1])
    return 1 -- Success
else
    return 0 -- Sold Out
end