Salesforce ↔ Marketo Sync Service

The existing sync was a cron script that fetched everything, pushed everything, and failed silently. When it hit a Salesforce API limit it dropped the batch; nobody found out until a sales rep noticed a missing lead days later. Rebuilding it meant handling two APIs with different rate-limit models, different field semantics and no shared identifier.

I built a standalone NestJS service around a RabbitMQ topology: one queue per direction, a delay queue for rate-limited retries with exponential backoff, and a dead-letter queue that keeps the original payload so any failure can be inspected and replayed by hand.

Every operation is idempotent on an external-ID key, so replays are safe. Field mapping and transformation rules live in configuration rather than code, which means marketing ops can add a custom field without a deployment. MongoDB stores sync state and a full payload history for debugging, and a small status endpoint surfaces queue depth and last-success timestamps to monitoring.

Silent data loss stopped being a category of incident: failures now surface within minutes and every one of them is replayable from stored payloads.