Have bosses understand the following situation: there was a cluster of 8 servers, each machine's memory is 128 g, all installed impalad and datanode, impala the client sends a complicated SQL to node 1, used in SQL table is very big, the operation of SQL when looked at each machine's memory usage, found that not all the machines are put on the memory usage, memory, are just three of the machine is close to overflowing, and the other a few seem to have no action, by rights, the cluster should have load balancing strategy, isn't it, what is want to know about the situation, I personally think that data files are just the three server storage, how the equilibrium distribution when writing data to the data file?