We’re seeing issues using :exponentially_longer where failures in downstream systems are causing a thundering herd effect. A great many jobs fail, they all get enqueued to try again in a static interval and bring down the downstream system yet again. This repeats until the retry limit is exhausted.
We’ve gotten around this using a custom
wait block but since exponential backoff with randomness is builtin to some adapters, like Sidekiq , I’m propsing to have it baked into to :exponentially_longer.
I submitted a PR here: https://github.com/rails/rails/pull/31872