Cloudera Manager reports that disks are full on some nodes of my Hadoop cluster but, checking manually the free space with df -h, every disk is almost half-empty.
Is it a bug or I need to issue some other commands to check the actual free space ? What command is Cloudera Manager using when doing health check ?
I solved the problem. The property dfs.datanode.du.reservedwas set to 100GB, so Hadoop could not use that amount of space (for each volume) for storing new HDFS blocks.