Integration · working demo

Break it on purpose

Records moving between three systems. Flip the switches to make the API fail, throttle, or die entirely, then watch retries, exponential backoff and the dead-letter queue make sure nothing is silently lost. That last part is what separates an integration that survives from one that quietly stops working.

Delivered 0 Retried 0

Pipeline

Execution log

Run a sync to see what happens.

Failure injection

Intermittent 500sRoughly a third of calls fail. Retries should absorb it.

Rate limiting429s with Retry-After. Backoff should pace the sync.

Destination offlineEverything fails all three attempts, straight to dead-letter.

High latencyCalls take 5× longer. Nothing breaks, it just crawls.

Dead-letter queue

Empty, nothing lost.

What's real here: a genuine async pipeline. Retries use real exponential backoff with real elapsed time, rate limiting actually paces requests, and the dead-letter queue holds real payloads you can replay. Turn the destination off, sync, turn it back on, and hit replay, the records complete. Nothing is ever silently dropped.

What's not: no network calls leave your browser.

Which two systems won't talk?

The connecting is the easy part. Making it survive outages, schema changes and rate limits is the job.

Describe your integration More integration work