01 / PLATFORM
Relay_
Relay accepts webhook traffic from third-party providers and guarantees delivery to internal consumers, even when those consumers are down. It absorbs bursts, retries with backoff, and keeps every delivery attempt auditable.
The hard part was not throughput but failure semantics. Making retries idempotent and replay safe meant treating every event as immutable and every delivery as a separate, observable attempt.
Highlights
- [ At-least-once delivery with idempotency keys per consumer ]
- [ Exponential backoff with a dead-letter queue and one-click replay ]
- [ OpenTelemetry traces spanning ingestion through delivery ]
- [ Horizontal scaling on partitioned Postgres with Redis for hot state ]