Ceph » rbd

Currently, ceph df shows two sets of statistics: global and pool.

For global, the output is generated from aggregating the data returned from the OS on each OSD host, so this show actual used/available.

For the pool data, the output is generated by ceph, and in fact shows provisioned space, not actual used space. This means for RBD pools with thinly provisioned files, the output can't be used for capacity management as the totals can be higher than the physical space on the disks.

Eg:

1. ceph df

GLOBAL:

SIZE      AVAIL     RAW USED     %RAW USED

1627M     906M      720M         44.29

POOLS:

NAME             ID     USED       %USED     OBJECTS

data             0      89183M     5.35      22529

metadata         1      6096K      0         22

rbd              2      813M       0.05      214

fedora           3      197M       0.01      61

.rgw             4      0          0         0

.rgw.gc          5      0          0         32

.rgw.control     6      0          0         8

.users.uid       7      444        0         1

.users           8      11         0         1

.users.swift     9      11         0         1

cac              10     759G       46.68     4294

The proposal is to rename the Used column to "Provisioned" and add a new column titled "Used" which shows real space used by a pool.