engelsanchez
commented
10 years ago Another unanswered question here: with the addition of the overload messages, special logic was added such that if a request failed and any of the vnodes returned an overload error, that overload error would be propagated to the client. With new known errors, there is the question of priority: what error to return if one vnode is overloaded and another hits the siblings maximum?
Except for overload messages, for which we hacked a workaround overloading the returned request id from a vnode reply. We should generalize that so that any error reasons are propagated back. For example, now the write can fail due to a violation of the max siblings or max size constraints.