[lustre-discuss] Recovering MDT failure

[Peter Grandi]

>>> On Thu, 27 Apr 2023 10:20:54 +0100, Peter Grandi
>>> <pg at lustre.list.sabi.co.uk> said:

>> - When I started this system I tried to backup MDT data
>> - without succes.

> [...] The ZFS version of MDT can do filesystem-level snapshots
> when mounted as 'zfs' instead of 'lustre'.

Just to be sure, even if this was already mentioned in a later
thread: even if the MDT is mounted as 'lustre' all 'zpool' and
'zfs' commands are still avaiable by using the zpool and
filesystem dataset names instead of the mountpoint, including
snapshotting (which should be read-only). 'zfs snapshot' and
'zfs send' do not require mounting the filesystem dataset, but
(always read-only) snapshots of the MDT mounted as 'lustre' can
be mounted as 'zfs' anyhow. My guess is that much more so for
'ldiskfs' the ZFS 'lustre' filesystem type is just a wrapper
around 'zfs'.