Retries & Replay

Repull automatically retries failed webhook deliveries with exponential backoff. You can also manually replay any delivery — up to 3 times per hour per delivery.

Retry Schedule

When a delivery fails, Repull retries it 7 times over 24 hours. Times are counted from the first attempt, so an endpoint that is down for a deploy, an outage or most of a day still receives the event:

AttemptAfter the first attemptWait since the previous attempt
1st retry1 minute1 minute
2nd retry5 minutes4 minutes
3rd retry30 minutes25 minutes
4th retry2 hours90 minutes
5th retry6 hours4 hours
6th retry12 hours6 hours
7th retry24 hours12 hours

Retries are sent within about a minute of these times. Every retry carries the same eventId (in the body and the X-Repull-Event-Id header) and a fresh X-Repull-Delivery-Id; X-Repull-Attempt counts up from 1. Every attempt, failed or not, is in the delivery log.

With several endpoints subscribed to the same event, each one is retried on its own: an endpoint that accepted the event is not sent it again, and one that failed keeps its schedule. If the 24-hour retry also fails, the delivery stays in the failed list and you can replay it from there. A successful manual replay ends the automatic retries for that endpoint.

What Triggers a Retry

A delivery is considered failed and will be retried if:

  • Your endpoint returns a non-2xx status code (e.g. 500, 502, 503)
  • Your endpoint does not respond within 10 seconds (request timeout)
  • The connection is refused or the DNS lookup fails

Return 200 quickly

Process webhook events asynchronously. Return a 200 status immediately, then handle the event in a background job. This prevents timeouts and duplicate deliveries.

Auto-Disable

Repull switches an endpoint off only when it keeps failing: 20 or more failed deliveries in a row, with no successful delivery for 24 hours. A short outage or a burst of timeouts never switches it off. Once it is off, re-enable it from the dashboard or with PATCH /v1/webhooks/{id} after fixing the receiver.

Viewing Failed Deliveries

List recent deliveries for a webhook subscription, including status, response code, and timestamps:

curl https://api.repull.dev/v1/webhooks/wh_abc123/deliveries?status=failure \
  -H "Authorization: Bearer sk_live_YOUR_KEY"

Manual Replay

Re-deliver any webhook delivery. The replayed delivery carries the exact same payload — and the same stable eventId — as the original. The path nests the delivery under the subscription it belongs to:

curl -X POST https://api.repull.dev/v1/webhooks/wh_abc123/deliveries/del_xyz789/replay \
  -H "Authorization: Bearer sk_live_YOUR_KEY"
{
  "id": "del_new456",
  "webhook_id": "wh_abc123",
  "original_delivery_id": "del_xyz789",
  "status": "success",
  "response_code": 200,
  "replayed_at": "2026-06-15T14:30:00Z",
  "replayNumber": 1,
  "replaysRemaining": 2,
  "nextWindowResetAt": "2026-06-15T15:30:00Z"
}

Every successful replay reports where you stand against the budget described below: replayNumber is which replay this was in the current window, replaysRemaining is how many are left in it, and nextWindowResetAt is when the window clears.

A delivery about a listing that is inactive now is not re-sent. The replay returns 403 listing_inactive with the listing in listing_ids; activate the listing, then replay again. See listing_inactive.

A replay is a new delivery, but the same event

Each replay is sent as a brand-new attempt: it gets a fresh X-Repull-Delivery-Id (the id in the response above). The logical event id is unchanged — it arrives in the X-Repull-Event-Id header. Always dedupe on eventId, never on the delivery id, or a replay will be processed twice.

Replay Limit

A delivery can be replayed at most 3 times per rolling 60-minute window. The window opens on the first replay and resets once it elapses, so a delivery you have not replayed for an hour starts fresh with all 3 again.

This is not a lifetime cap. An event whose endpoint was broken last week can still be replayed today — there is no age limit on replays.

The budget belongs to the original delivery

A replay produces a new delivery, and that new delivery can be named in a replay call of its own. It draws on the same budget: replaying a replay counts against the original delivery's 3 per hour. Chaining replays is not a way around the limit.

A delivery that already succeeded is refused

If your endpoint returned 2xx for a delivery, replaying it would only hand you a duplicate of an event you already processed. The call is refused with 409 delivery_already_succeeded. See delivery_already_succeeded.

To send one anyway — you dropped the event on your side, say, or you are rebuilding state — pass force:

curl -X POST https://api.repull.dev/v1/webhooks/wh_abc123/deliveries/del_xyz789/replay \
  -H "Authorization: Bearer sk_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "force": true }'

A forced replay still counts against the 3-per-hour budget.

When you are over the limit

The 4th replay inside the window returns 409 replay_limit_reached with a Retry-After header in seconds, and the same wait expressed as an ISO-8601 timestamp in the body:

HTTP/1.1 409 Conflict
Retry-After: 2280

{
  "error": {
    "code": "replay_limit_reached",
    "message": "Delivery del_xyz789 has been replayed 3 times in the last hour.",
    "docs_url": "https://repull.dev/docs/errors/replay_limit_reached",
    "replays_made": 3,
    "replay_limit": 3,
    "next_replay_allowed_at": "2026-06-15T15:30:00Z"
  }
}

Why 409 and not 429

The conflict is with the state of that one delivery, not with the pace of your requests. Your key is not throttled: other endpoints keep working, and replays of other deliveries go through normally. Do not route this into the same backoff path as rate_limited.

Before You Spend a Replay

Failed deliveries are already retried automatically on the schedule above. A manual replay is for after you have changed something — if the same delivery keeps failing, replaying it again will not fix it, because the receiver is what is broken. Check, in this order:

  • Your endpoint returns a 2xx within the 10-second delivery timeout. Acknowledge first, process in a background job.
  • Your signature check is not rejecting valid requests. See Verify Signatures.
  • GET /v1/webhooks/{id}/deliveries records the response body and status code your endpoint returned for each failure — usually the fastest way to see what it is actually complaining about.
  • Send yourself a test event with POST /v1/webhooks/{id}/test/{event_type} to confirm the endpoint accepts events, before you spend replays on real ones.
# What did the endpoint actually say?
curl "https://api.repull.dev/v1/webhooks/wh_abc123/deliveries?status=failure" \
  -H "Authorization: Bearer sk_live_YOUR_KEY"

# Confirm the fix with a test event — this does not touch the replay budget
curl -X POST https://api.repull.dev/v1/webhooks/wh_abc123/test/reservation.created \
  -H "Authorization: Bearer sk_live_YOUR_KEY"

# Only then replay the real delivery
curl -X POST https://api.repull.dev/v1/webhooks/wh_abc123/deliveries/del_xyz789/replay \
  -H "Authorization: Bearer sk_live_YOUR_KEY"
AI