New Disk in Data Node

[#1]

Hello,

We have a 12 node cluster with 4 data drives in them each. We added 1 additional disk to each data node but are having a hard time getting data to be placed on these additional drives. For example:

node01 - 

/hdfs/01 - 1.4 TiB/1.8 TiB

/hdfs/02 - 1.5 TiB/1.8 TiB

/hdfs/03 - 1.5 TiB/1.8 TiB

/hdfs/04 - 1.5 TiB/1.8 TiB

/hdfs/05 - 18 GiB/1.8 TiB

node03 - 

/hdfs/01 - 1.5 TiB/1.8 TiB

/hdfs/02 - 1.5 TiB/1.8 TiB

/hdfs/03 - 1.5 TiB/1.8 TiB

/hdfs/04 - 1.5 TiB/1.8 TiB

/hdfs/05 - 181.9 GiB/1.8 TiB

We've run the rebalancer for a few weeks now. Would it be easier to just wipe one data node at a time and introduce it back into the cluster?

Thanks,

Eric

[#2 · Accepted solution]

The HDFS Balancer only balances blocks between DataNodes.   It does not to any balancing on individual DataNodes between drives.

You can set the DataNode Volume Choosing Policy (dfs.datanode.fsdataset.volume.choosing.policy) to Available Space (org.apache.hadoop.hdfs.server.datanode.fsdataset.AvailableSpaceVolumeChoosingPolicy).  This will cause the DataNodes to write new blocks to the drive with the most space available.   It does not affect blocks that have already been written.

For your question about wiping one datanode at a time, it would be better to decommission and then recommission a node.  With a replication factor of 3 you may perform this action on 2 nodes at a time.

[#3]

Thanks for the quick reply. Much appreciated.