# Retries and redelivery

The retry schedule, when a delivery is held instead of retried, how endpoint health changes, and how to resend deliveries by hand or in bulk.


A delivery succeeds when your endpoint answers any `2xx` status within 10 seconds. Anything else, including a timeout, a refused connection or a redirect, is a failed attempt. What happens next depends on the answer.

## The default schedule

After a failed attempt the delivery returns to `pending` with a `next_attempt_at`. The waits after each failure are 30 seconds, 2 minutes, 10 minutes, 30 minutes, 1 hour, 2 hours, 4 hours and 8 hours; from then on, every 12 hours. Each wait gets up to 10% random jitter so that retries from many deliveries do not arrive together. A delivery stops as `failed` after 15 attempts or once it is 72 hours old, whichever comes first. The 72-hour limit is a server setting that the operator of your installation can change; the schedule and the 15 attempts are fixed.

## Which answers are retried

| Your answer | What happens |
| :--- | :--- |
| `2xx` | `succeeded`. |
| `5xx`, `408`, `425`, `429`, timeout or connection error | Retried on the schedule above. |
| Any other `4xx` | Sent twice in total, then held as `paused` until you act. |
| `410 Gone` | The endpoint is paused automatically and its queued deliveries are held. |

A held delivery is never dropped while it waits, but it still turns `failed` when it reaches the maximum age.

## Endpoint health

Each endpoint carries a `health` field. It moves to `degraded` after three consecutive failed attempts, to `failing` once attempts have failed for 24 hours, and to `paused_automatically` after 72 hours of failures. Pausing it yourself sets `paused_by_owner`. Each change caused by a delivery attempt is written to the audit trail as `developer.webhook.health_changed`; pausing or resuming by hand is recorded as `developer.webhook.updated` or `developer.webhook.resumed`. One successful attempt resets the count to zero.

Deliveries about the same record stay in order: while an earlier event about an invoice is undelivered, a later one about that invoice waits behind it.

## Resend one delivery

```bash
curl -sS -X POST "$BUSINESS_API/developer/deliveries/dlv_2kP8aQ5mZr1NcV7xT3bLwE9fGyU/retry" \
  -H "Authorization: Bearer $API_KEY"
```

The `202` response shows the delivery due again with a fresh attempt budget:

```json
{
  "id": "dlv_2kP8aQ5mZr1NcV7xT3bLwE9fGyU",
  "tenant_id": "ten_2jR5wQ8nLm3KxT9vB6cYpZ1dHgF",
  "webhook_id": "whk_2kP7vR3nXq9LmT4bW8cYzF1dHsJ",
  "url": "https://erp.example.test/hooks/billing",
  "event_id": "wev_2kP9bR7nWu2JfT5xQ1mKhB8cVzD",
  "event_type": "invoice.paid",
  "status": "pending",
  "attempts": 15,
  "max_attempts": 30,
  "next_attempt_at": "2026-11-19T10:31:05.204118Z",
  "response_status": 503,
  "duration_ms": 412,
  "created_at": "2026-11-16T10:12:44.930551Z",
  "sequence": 4
}
```

It answers `409` while the delivery is being attempted, or when its webhook was deleted. In the dashboard the same action is **Resend** on the delivery page.

## Resend in bulk

`POST $BUSINESS_API/developer/webhooks/{id}/redeliveries` puts an endpoint's `failed` and `paused` deliveries back in the queue for a window of at most 7 days (`from` and `to`, RFC 3339). You can narrow it with `types` and `statuses` (which also accepts `cancelled`). Send `"dry_run": true` first: the `200` response gives the count in `requeued` and changes nothing. The real call answers `201` with a `wrd_` identifier. The dashboard offers this as **Resend failed** on the webhook page.

Retries, resends and bulk redeliveries reuse the same delivery, so `webhook-id` stays the same every time.

## Next steps

- [Make your receiver ignore repeated deliveries](/docs/webhooks/best-practices)
- [Check the delivery log while testing](/docs/webhooks/testing)
- [How errors such as 409 are reported](/docs/getting-started/errors)

