@Valdanito thanks so much for suggesting --rebalance_interval trick.
- I completely rebuilt my cluster with setting --rebalance_interval 1200000h when starting zero’s
- My cluster is looking so beautiful
- Although, predicates are bit imbalanced
For now, only my biggest unresolved issues is, bulk uploading more than 350GB (compressed size) data. Below is the command I run, let me know if you have similar tricks 
dgraph bulk -f /coldstart/upload/pending_predicates -s /coldstart/upload/rdf_schema/patient.rdf --replace_out --num_go_routines=20 --reducers=7 --format=rdf --store_xids --map_shards=14 --reduce_shards=7