On Wed, May 01, 2019 at 11:06:37AM -0500, Andrew Schoen wrote:
I'm aware that one could work around this by creating the LVM setup that I want.
I think this is a bad approach though since every deployment tool has to
implement its own LVM handling code. Imho the right place for this is exactly
ceph-volume.
I've been thinking about porting some of the `batch` lv creation functionality
to the `create` subcommand. I think it'd be nice to be able to pass a
raw device to any of the
available flags in `create` and if that device has already been used
as a pv and a vg exists then
that vg would be reused, otherwise a new pv and/or vg would be
created. This is beneficial because it
gives the control that users seem to wish `batch` had, like `create`
already does. Would that solve this problem for you?
I think there's a lot merit to this approach! This could also neatly solve some
other warty situations, e.g. when batch can create wal/db LVs but not the
respective block LVs there situations where a rollback is not (easily) possible.
That leaves behind some unused LVs but no OSDs.
I'd be hesitant to change `batch` here. Originally we went down the
path of having one vg per pv for
db/wal but it added tons of complexity to the code to manage that.
With fairly simple workarounds existing for
`batch` plus the availability of the `create` subcommand we felt it
wasn't worth the potential maintenance and
complexity burden to put this sort of functionality in `batch`.
Best,
Andrew
--
Jan Fajerski
Engineer Enterprise Storage
SUSE Linux GmbH, GF: Felix Imendörffer, Mary Higgins, Sri Rasiah
HRB 21284 (AG Nürnberg)
