If you look at catalogd memz, e.g. at localhost:25020/memz, it has a bogus breakdown. The catalogd does not use the MemTracker infrastructure since the vast majority of it's memory consumption is in the JVM.
Reported by alanj_impala_5a78
Update: It is the same for statestored as well and apart from "breakdown" section, the "Memory consumption / limit" part is also redundant. Should remove both parts from the memz pages of both daemons