Appearance
Error handling
The same rules for every capability. Nothing is lost when you're briefly unavailable: we retry in the same order, and because everything is an upsert, a retry never does anything twice.
| When you answer | What we do |
|---|---|
2xx | Done. |
429 | We wait for as long as Retry-After says (in seconds) and continue where we were. |
5xx, a timeout, no connection | You're unavailable. We retry later with increasing pauses, in the same order. We wait up to 30 seconds for an answer. |
401, 403 | The key no longer works. We stop calling you and ask the organiser for a new key. |
404 on one record | It doesn't exist (any more), for example GET /v1/orders/{id} for an unknown order. We skip it. |
422 | Our request broke the contract, or you can't accept it. We don't retry it unchanged; we show your message to the organiser and to our engineers. |
Any other 4xx | A rejected request. We don't retry it unchanged, and show your message to the organiser. |
Error messages
With any 4xx you can send a message, which we show to the organiser as it is. It is optional.
json
{
"message": "Product type tt_9 does not belong to event evt_123."
}Invalid records
When one record in a page breaks the contract, we skip it, report it with the path of the field involved, and continue with the rest of the page. A page that isn't a page at all fails as a whole.
Webhook responses
Our webhook URL answers 2xx as soon as it has your call. Anything else is a short outage on our side: retry later if you can, and polling covers the rest.