It would be nice if the retry times for integration actions had some noise added to the retries. If the customer triggered 500 actions at once, we get 500 actions at once. We can’t handle those due to monday’s API limits. So we reject them with a 429 so they can retry when we can actually process them. Since they all started at the same time - their retries are at the same time.
It would be nice if retries were randomized by like 10s.
Alternately, honor our retry-after headers. Then we could randomize. We could also count rejected events, and when the number rejected exceeds what we could process in the next minute we can delay even further, and so on. This way we’re not dealing with requests we can’t process easily.