← Back to selected work
02 / BACKEND RELIABILITY LABLocal demo

RelayBox

Every event. With a way back.

A local webhook delivery lab that makes the journey from queued to received visible—even when the receiver fails along the way.

PROJECT TYPE

Backend reliability lab

BUILT WITH

Node.js 24 · SQLite / WAL · JavaScript modules · HTML / CSS

STATUS

Local demo

Application screenshot · sample/demo data
THE CHALLENGE

Where the simple version breaks.

A receiver can be offline when an event arrives. It can also accept a request but respond too late. A reliable delivery workflow needs more than a send button: it needs persistence, bounded retries, and a history that explains what happened.

THE WORKFLOW

A clear path through the task.

Events are stored before a worker sends signed requests to built-in local receivers. The dashboard exposes delivery status, attempts, response codes, timing, verified signatures, and replay relationships.

ENGINEERING DECISIONS

The details that make it hold.

01

Persist before attempting delivery.

SQLite stores the queue, configuration, and attempt history. Transactional claims and worker leases allow interrupted deliveries to recover after a restart.

02

Expect failure and bound the response.

Four attempts, exponential delays of 2, 4, and 8 seconds, and a 3-second timeout. Exhausted deliveries enter a dead-letter state for inspection.

03

Make retries and replay explainable.

Each attempt uses a timestamped HMAC-SHA256 signature. A replay creates a new linked delivery, preserving the original history. Delivery is at least once; receivers still need idempotent processing.

VERIFICATION

Behavior, checked.

The recorded full local suite passed 13 tests, including real HTTP delivery and restart persistence. Twenty concurrent requests using the same idempotency key produced one event. The browser demo showed HTTP 503, 503, then 200 with verified signatures.

CURRENT SCOPE

Where this project stands.

A single-user local lab that sends only to three built-in receivers. It has no public-user authentication or arbitrary external destinations. Public deployment needs authentication, usage limits, and operational controls. Demo counts are not production reliability benchmarks.

NEXT PROJECT

Renew↗

Less surprise. More living.

Expanded project screenshot