I looked up RBD disk space usage, but found different statistics from Ceph and the host which mounts the disk.
From Ceph:
$ rbd -p rbd du
NAME PROVISIONED USED
kubernetes-dynamic-pvc-13a2d932-6be0-11e9-b53a-0a580a800339 40GiB 37.8GiBFrom the host which mounts the disk
$ df -h
Filesystem Size Used Available Use% Mounted on
/dev/rbd0 39.2G 26.6G 10.6G 72% /dataHow could I explain the difference?
You can check mount options of /dev/rbd0 device. There should be no 'discard' option. Without that option filesystem cannot report to Ceph about reclaimed space. So Ceph has no idea how much space is actually occupied on rbd volume. This is not a big problem and can be safely ignored. You can rely on stats reported by kubelet.