I have 2 questions regarding disk usage for hdfs.
1. what happens if the disk on which the folder specified for "dfs.data.dir" is
full?
2. I know one can specify "dfs.datanode.du.reserved" to reserve certain amount
of disk space for non-dfs use. But is this expensive? Will data node have to do
a "du"-like operation before each block-write request or what actual operations
will a data node perform if this setting is activated? Also I didn't find
"dfs.datanode.du.pct" in current hadoop documentation. Is it still effective?
By the way, I am using cloudera 0.20.2+228-1 package.
Best regards,
Michael