I need to run through some server maintenance on my data nodes, including a
reboot. My splitlogs, though, only seem to have a replication factor of 1
(when a data nodes is taken offline, I sometimes have missing blocks for
them). I know I can decommission data nodes with the exclude.dfs file, but
this takes many days to complete, and seems like overkill for a quick
reboot.

Is there a better way to go about this? It seems like splitlogs should have
a replication factor of at least 2, in case a server dies (which would
allow me to more safely reboot a server). I don't know the performance hit
this would cause, though.

Or maybe this is more of an HDFS question, and there's a way to tell a data
node to only shed it's replication=1 blocks.

Any thoughts would be appreciated.

Thanks,
Patrick

Reply via email to