Fan-out Routing
Fan-out Routing delivers one inbound webhook to many destinations at the same time. Each destination has its own Third-Party Model, its own filters, and its own retry behavior. A failure or outage in one destination has zero effect on any other — no shared queue, no shared rate limit, no shared retry storm.
Why use Fan-out Routing?
Most webhook pipelines start with one destination. By month three, you need three: your CRM, your warehouse, and your analytics tool. If you fan out in application code, three things go wrong:
- Failure couples destinations. A slow analytics API holds up the request, the warehouse times out, and you start dropping deliveries to everything.
- Retry logic gets duplicated and inconsistent. Each destination needs its own backoff, its own circuit breaker, its own failure handling — and you re-implement all of it for each one.
- Adding a destination means writing code. Marketing wants Klaviyo. Now it’s a deploy.
Fanning out at the gateway makes destinations independent by construction. Each one is its own pipeline. Adding a destination is configuration, not code. And one destination going down stays one destination going down.
How it works
When a webhook arrives, it’s validated against the Golden Model once. Then each configured destination runs its own filter, applies its own Third-Party Model mapping, and delivers through its own retry pipeline.
graph TD
S["Inbound Webhook"] --> G["Golden Model\n(validated once)"]
G --> F1{"Filter: Analytics"}
G --> F2{"Filter: Fulfillment (Dev)"}
G --> F3{"Filter: Fulfillment (Prod)"}
F1 -->|pass| DM1["Third-Party Model"]
F2 -->|pass| DM2["Third-Party Model"]
F3 -->|pass| DM3["Third-Party Model"]
DM1 --> D1["Analytics"]
DM2 --> D2["Fulfillment (Dev)"]
DM3 --> D3["Fulfillment (Prod)"]
style G fill:#a855f7,color:#fff
style D1 fill:#7c3aed,color:#fff
style D2 fill:#7c3aed,color:#fff
style D3 fill:#7c3aed,color:#fff
Three deliveries from one inbound webhook. If Fulfillment (Prod) is down, only its circuit breaker trips. Analytics keeps flowing. Fulfillment (Dev) keeps flowing. Attempts are bounded; when they run out, the production delivery is permanently failed for that destination and shows in its delivery status. When the production endpoint recovers, you replay the event — the original payload is re-sent exactly as the vendor sent it.
Example
A slashbin customer’s fulfillment-routing pipeline ingests order webhooks from a single Shopify source. The fan-out delivers to three destinations:
- Analytics — every order, full Golden Model payload, fire-and-forget.
- Fulfillment (Dev) — only orders tagged
staging, mapped to the warehouse’s test API. - Fulfillment (Prod) — production orders only (filter:
tags does not contain "staging"), mapped to the warehouse’s production API.
Each destination has its own Third-Party Model and its own filter. Last quarter the production warehouse had a four-hour API outage. Analytics and Dev fulfillment continued without interruption. When production recovered, the customer replayed the events whose production deliveries had failed and every missed order reached production. A replay re-sends the event to every active matching destination, including Analytics, which had already received it — so each consumer has to tolerate seeing an event again.
Related concepts
- Golden Model — the canonical schema validated once before fan-out, so every destination sees consistent data.
- Third-Party Model — the target-shaped schema each destination uses, derived from the Golden Model.
- What’s the best way to replay webhooks? — how a replay re-sends the original payload to recover from outages.