The data sets the Dgraph queries are stored in databases. For high availability, these can be stored on HDFS/MapR-FS or a shared NFS. They can also be stored on the local disk for a non-HA option.
The location you choose determines the database requirements, as well as where the Dgraph will be installed and its behavior.
Regardless of where you put your Dgraph databases, you must increase the maximum numbers of open file descriptors and processes on all nodes in your cluster (including Hadoop nodes), or the Dgraph may crash during processing.