As you can see I tried to use roundrobin and most available to make the node in cluster being balance.
I use a spark job read from current data in Alluxio and do some summary then write back to Alluxio without replace a old one.
It will be nice if you guys provide some notice about which file and folder is being missed.
cluster include 5 nodes:
node 2,3,4,5: 20GB
node1 also a master.