Created on 11-04-2015 12:26 PM - edited 09-16-2022 02:47 AM
Hi List,
I have disk space limitation on couple of my datanodes (also have regionserver and nodemanager roles on them). It is on a public cloud, I already have the largest disk and they would not give me more disk.
These machines are also running other high-priority processes (other than CDH processes), which needs disk space. So I want to free up space by taking hbase/hdfs off it and distribute the data to other nodes in the cluster.
I have decommissioned the nodes and deleted the roles from the hosts but that does not free up /dfs directory on the hosts (still taking most of my disk space).
Is it OK to directly delete the /dfs directory from the nodes, or make it empty? I want the flexibility of adding them back to the cluster at some point of time (may be once I have more disk), if I need to. I am hoping I am not the first one in this situation, any suggestions/work-arounds would be appreciated.
Thanks,
Akshay
Created 11-09-2015 02:27 PM
Happy to say that our in documentation we have noted that when a DataNode is decommissioned, the data blocks are not removed from the storage directories. You must delete the data manually.
Created 11-09-2015 07:28 AM
Bump up.
Sorry still confused about it.
Created 11-09-2015 02:27 PM
Happy to say that our in documentation we have noted that when a DataNode is decommissioned, the data blocks are not removed from the storage directories. You must delete the data manually.
Created 11-10-2015 05:27 AM
Thanks Michalis.
I believe if I have to add this back to the cluster, those directories (of course based on configurations) would be created again.