The $joinPlanCacheStats stage currently iterates over the entire cache and serializes every entry to BSON before returning vector<BSONObj> over the MongoProcessInterface. For the EAP, this is ok as the size of the cache is controlled by a query knob so the memory used by this stage is capped. For GA, we need to implement some hardening to ensure that this stage does not use a large amount of memory without removing its diagnostic value.