Webhook retry behavior on failed deliveries

Question

Seeing inconsistent delivery on our webhook endpoint. Some events arrive 3-4 times. Need to understand:

What's the retry policy? Exponential backoff or fixed intervals?
How many attempts before FieldPulse gives up?
Is there a dead letter queue or visibility into failed deliveries?

Currently returning 500 on purpose for testing — seeing retries come in at weird intervals. No docs on this.

Also: what's the idempotency key situation? Need to dedupe on our end.

Eli Torres (Staff) · Accepted Answer

Carlos — yeah, the retry behavior isn't super well documented yet (we're fixing that). Here's what's happening: Current retry policy: Exponential backoff: 1s, 2s, 4s, 8s, 16s, then every 60s Max attempts: 24 hours worth of retries, then the event is dropped No dead letter queue today — we just log and move on The "weird intervals" you're seeing are probably the exponential backoff plus some jitter we add to prevent thundering herd. The jitter is ±20% of the calculated delay. Idempotency: Each delivery attempt has a unique X-FieldPulse-Delivery-Id header, but that's per-attempt, not per-event. The event itself has event.id in the payload — that's what you want to dedupe on. Example payload structure: {
 "event": {
 "id": "evt_abc123",
 "type": "work_order.status_changed",
 "created_at": "2025-08-18T10:20:00Z"
 },
 "data": { ... }
} Store evt_abc123 with a TTL and ignore duplicates. We don't guarantee exactly-once delivery. Worth noting: if your endpoint is consistently failing (5xx or timeouts > 30s), we do start backing off the connection pool for your endpoint, which can delay initial delivery attempts. So that 500 test might be affecting observed behavior. We're tracking a proper delivery dashboard internally — no ETA but it's on the roadmap. For now, email me your webhook ID if you want me to pull recent delivery attempts from our logs.

Devon Marsh · Answer

Heads up on the dedupe approach — we learned this the hard way. The event.id is stable across retries for the same event, but if a webhook subscription gets deleted and recreated (even with the same URL), you'll see new event IDs for the same underlying change. Rare edge case but burned us during a migration. What we ended up doing: composite key of event.id + hash of the payload's data object. Extra protection against accidental double-processing if someone clones a subscription. YMMV but saving you the headache we had.