05-14-2018 02:37 PM
In the Cloudera Reference Architecture for Bare Metal deployments (2018-04-30), there is a section on Filesystems which says - In RHEL/Centos the LVM should never be used for data drives. It can be beneficial to keep LVM enabled on the OS drives. Then it goes on to say -
Disk Mount Naming Convention: For ease of administration, it is recommended to mount all of the disks on the DataNode machines with a naming pattern, for example:
/data1 /data2 /data3 /data4 /data5 /data6
In a typical production Datanode it is common to have 16-24 disk drives with each disk 1-4TB for total 64-100TB per datanode. Does the above mean that we have to define 24 file system /data1,/data2..../data24 without LVM. And then we have to specify all these 24 mounts during CDH install. If you have say 10 datanodes then you will have to define 24*10=240 file mounts. This sound bit crazy to me if we cant combine reasonable number of disks together in LVM and have say about 5 file mounts instead of 24 mounts per server for ease of administration. Appreciate any comments from the CDH install experts. Thanks!