Stack: Redis (advisory locks (leases)), PostgreSQL (truth), FastAPI (API)
Focus: Concurrency correctness, idempotency, atomic multi-seat booking, load-tested under race conditions.

This project is intentionally backend-heavy. There is no UI, no auth, and no payments.
The goal is to solve seat allocation under concurrency correctly and prove it with load tests.

The system is split into coordination and truth layers:

• Redis is used only for advisory locking with TTL.
• PostgreSQL is the single source of truth.
• FastAPI exposes HTTP endpoints.
• Docker Compose runs everything locally in isolated containers.

Redis helps reduce contention.
Postgres is the source of truth.

The Redis stage absorbs high-concurrency bursts and filters contention early, preventing concurrent race condition-causing requests from overwhelming the database. The PostgreSQL stage is the final authority, enforcing durability and correctness using transactional guarantees and row-level locks.

ALTER TABLE bookings
ADD CONSTRAINT uq_booking_request_seat
UNIQUE (request_id, seat_id);

This allows:

• One request_id → multiple seats
• Safe retries
• No duplicate seat booking

POST /seat-lock
{
  show_id,
  seat_ids,
  user_id
}

• Redis keys: lock:{show_id}:{seat_id}
• TTL-based
• Failure returns 409

POST /booking/confirm
{
  show_id,
  seat_ids,
  user_id,
  request_id
}

• Atomic multi-seat booking

• No partial commits

• No double booking

#	Test Name	What It Validates	Seats	Users	Ramp-up	Runtime	TTL	Screenshot
1	Lock Contention	Mutual exclusion	A1	50	1/sec	2m	120s
2	Lock TTL Expiry	Lock auto-release	A1	50	1/sec	1m	5s
3	High Ramp-up Lock Race	Atomic locking	A1	50	10/sec	1m	5s
4	Confirm Race (Single Seat)	DB truth beats Redis	A1	20	5/sec	1m	1s
5	Confirm Race (Multi-seat)	Atomic commit	A2,A3	20	5/sec	1m	1s