Impala's support for Llama, which adapts Yarn for low-latency query engines, has long since bitrotted, and never worked very well. I propose removing it from the next version of Impala - I don't know of any deployments relying on its use.
The initial patch is here: http://gerrit.cloudera.org:8080/4445 If there's any discussion to be had, let's do it here. Thanks! Henry