According to a conversation with adar, the usual way to get a trace for a slow RPC in Kudu is to set a deadline.
In Impala we don't use RPC deadlines, so would like to get RPC traces when --rpc_duration_too_long_ms is exceeded. In this case there is already logging, but only metrics are included: https://github.com/cloudera/kudu/blob/978cf39/src/kudu/rpc/rpcz_store.cc#L268
We should log the trace in this case. Adar pointed out that we should not log a redundant trace if it was already logged for the RPC deadline.