02-21-2017 12:24 PM
Need to perform hardware maintenance(not disk related) on a datanode.
- Is decommisioning a datanode is required if maintenance is only for a single datanode?(Decommission is taking ~1hour)
- If multiple datanodes on the maintenance schedule, decommissioning datanodes before maintenance is required?
Which is the ideal approach?
Appreciate your thoughts on this.
02-21-2017 10:40 PM
It’s a bad idea to stop single or multiple Datanode daemons or shutdown .
could you please let me know how are you performing the decommission ?
are you performing the decommission by adding the host address in the dfs.host exclude file ?
after peforming hardware shutdown , it recommend to run the hdfs balancer
02-22-2017 09:35 AM
Thank you @csguna for the reply!
Decommissiong the node is through cloudera manager. It used to take ~1hr for block replication from that node.
Instead planning to stop roles on the host(number of hosts is < 3 for maintenance) and shutdown without decommissioning(something like server crash). Hoping under replicated blocks from that maintenance hosts will take care periodically by namenode without waiting for decommission process to complete as it is resilient to handle the missing blocks. Maintenance is not related to disk.
Is this approach is ideal if number of maintenance datanode is less than 3 as rep factor is 3?
Please advise your thoughts.