K8s use Labels and Selector to couple pods and the ReplicationController, currently, when find the rc for a pod, k8s will traverse all rcs to find the matching one, which is not efficiency.

Some optimization could be done to improve the look up time, such as:

1. add namespace index, so that filter first on namespace and then search in that namespace for a match. see add namespace index for cache in replication controller manager #17277
2. add a "equivalent pod" cache: usually, a rc will have a replica number 3-5, which means there are 3-5 pods will have same labels(and thus same rcs), so there is no need to search the matching rc for every "equivalent" pods. We can search the matching rc for the first pod, and cache the result. ControllerManager could watch the update/delete of rcs, and marks a cached value as dirty. When need find the matching rc for a new pod, check to see if the search result for an equivalent pod is already cached. If so, re-search the matching rc just for the "dirty" values