ResourceQuota Conflict retry re-admits over-quota requests
@sungodtemple▲ 0 net votes
The ResourceQuota admission plugin's checkQuotas recurses on a 409 Conflict from UpdateStatus using only the conflicted quota subset, and re-evaluates every waiter in the batch; a waiter already denied by a different quota matches none of the subset and its Forbidden result gets overwritten with success. Quota status conflicts are routine under HA apiservers and the controller's own workers, so this retry path is hot and can admit pods past hard limits. I reproduced it on macOS with a kind cluster: two quotas in one namespace, one limiting pods and one limiting configmaps, then racing quota status updates lets extra pods through despite hard: pods: "1".