Hi Silvio,
Hadoop client libraries can be installed on any edge node or master / slave are an entry point for clients accessing services in your cluster. If this is a dev test environment, I typically install the clients across several nodes, both inside and on the edge. It can never hurt to have multiple dev test hadoop client paths. I typically steer clear from slave nodes, and pick master nodes that have less workload or are not part of HA configuration requirements. <span style=”line-height: 1.5;”>If it is production, you want to designate nodes as Hadoop Client entry points to your cluster and they may only service this purpose. </span>
Good luck,
Mark