The Runtime Theory
Backend EngineeringIn production

API Idempotency Keys

Recording in progress
#idempotency#apis#reliability

When a client retries a failed request, the server cannot tell a duplicate from a new request — unless the client sends an idempotency key. This video shows how the key is stored, how a unique constraint deduplicates retries at the database level, and what happens when two retries race through the handler at the same time.

Topics covered:

  • The problem: retries create duplicate side effects
  • The key: a client-generated string stored with the request
  • Dedupe tables and unique constraints at the database level
  • First-write wins: locking and the concurrent retry race
  • Key expiry and the replay window
  • Payment APIs and why Stripe made idempotency keys standard

Related articles

More in Backend Engineering

In production
backend

Pagination Strategies

Offset, keyset, and cursor pagination compared by database cost, consistency, and behavior when rows are inserted mid-page.

Details
In production
backend

Caching for the Service Layer

Service-layer caching done right — cache keys, invalidation, TTLs, and the stampede problem that brings services down.

Details
In production
backend

Webhooks vs Polling

Webhooks and polling compared at the delivery level — delivery guarantees, retries, and when push is worse than pull.

Details
In production
backend

Database Transactions in Practice

ACID in a real database — write-ahead logs, locks, undo and redo, plus isolation levels and what they change about reads.

Details
In production
backend

Rate Limiting Algorithms, Visualized

Token buckets, leaky buckets, and sliding windows compared by memory, precision, and what they do under burst load.

Details
In production
backend

Background Job Processing

Queues, workers, and retries — how background jobs actually execute, from enqueue to acknowledgment, and why they fail.

Details
In production
backend

REST APIs from the Inside

What happens when a REST API processes a request — routing, deserialization, validation, handlers, and response building end to end.

Details
In production
backend

Message Queues in Practice

Brokers, producers, and consumers — what a queue does at the protocol level and why it decouples services from availability.

Details
In production
backend

Connection Pooling Explained

What a connection pool actually does — the handshake cost it hides, checkout and release mechanics, and how pool size changes latency.

Details

Depth, delivered weekly

One technical dispatch a week — articles and episode notes before they go public.

One technical dispatch per week. No noise.