We have a Cluster running CDH 5.15.1 on 150+ nodes. Cluster nodes storage capacity varies (Mix of various node sizes, 10 TB, 13 TB, 25 TB, 48 TB).
Nodes are connected with a 10G network and presently running with a physical topology of Rack awareness.
Issue is, while few racks have 18 servers others are just having 1 or 2 servers on them. we have been advised to move away from this nonuniform rack awareness topology to a Logical Rack awareness topology which will help us put together UNIFORM no. of nodes on each rack.
I would like to understand, if the rack awareness topology is changed, would HDFS kick-off immediate Block movement or the new topology will only be effective for NEW blocks written on the cluster.
... View more