I’m building a real-time flash-sale ticketing queue. I’m choosing a containerized Go/Redis monolith over an AWS Serverless event-driven architecture.

[Trideep Dutta]

Hey everyone, I’m finalizing the architecture for a high-concurrency ticket reservation system (think concert tickets or limited sneaker drops).
The Constraints:

• Traffic: We expect spikes of up to 50,000 concurrent users hitting the "Buy" button within the first 30 seconds of a drop.
• Inventory: Only 5,000 items available. Overbooking is absolutely not an option.
• Latency: Needs to feel instant to the user, either giving them a confirmed lock on the item for 5 minutes to complete checkout, or an instant "Sold Out."

The Proposed Architecture (What I'm choosing):
I'm going with a vertically scaled, containerized Go API talking directly to a massive Redis instance, with PostgreSQL acting only as the final persistent truth after checkout.

• All inventory is loaded into Redis before the sale starts.
• The Go backend handles the queue and uses Redis atomic operations (DECR and Lua scripts) to lock inventory.
• Postgres only gets written to once the payment actually clears.

The Rejected Architecture (What I'm avoiding):
I explicitly rejected the popular "modern" approach of Serverless (AWS Lambda + API Gateway + SQS + DynamoDB).

• Why I hate it for this: Lambda cold starts and API Gateway latency add too much overhead when 50,000 users hit it at exactly 12:00:00. Plus, provisioning DynamoDB write capacity for a traffic spike that lasts literally 30 seconds feels like a complete waste of money and a recipe for throttling disasters.

The Ask: Destroy My Design
I know a single beefy Redis instance is technically a single point of failure, but I feel like keeping it simple is safer than managing a massive distributed event-driven mess.
As a Senior Engineer, tear this apart.

1. Where is my blind spot?
2. When this inevitably crashes under load, what is the exact component that catches fire first?
3. Prove to me why I should have gone with a distributed/serverless stack instead.

Roast away