Fix #6652: LVM Autoactivation Missing for Ceph OSD LVs - v2
===========================================================
In short: When creating an OSD via the API, the logical volumes backing
the OSD's DB and WAL do not have autoactivation enabled. Ceph requires
autoactivation on LVs, as it otherwise never activates them directly
itself. Fix this by setting autoactivation when creating those LVs as
well as providing a helper script that enables autoactivation for them
during an upgrade.
Notable Changes
---------------
- Add missing replacement for `LVMPlugin::lvcreate()` helper in
OSD.pm (thanks Fabian!)
- Drastically limit the scope of the helper script that runs during
upgrades, most of which was discussed off-list with Fabian as well
(thanks!). In particular:
- No longer activate matched LVs
- No longer try to bring up OSDs on the node
- Limit enabling autoactivation to LVs which back an OSD WAL or OSD DB
* previously, LVs that back OSD devices with type "block" were also
matched
- Limit the output of the helper script, most of which was also
discussed off-list with Fabian (thanks again!)
- The script is now completely silent unless an error is encountered
- An invoked command (ceph-volume / lvs / lvchange) fails, the
captured stderr of that command is dumped before exiting
- Otherwise, stdout is processed (or swallowed) and stderr is
suppressed in order to not worry any users with spurious /
unwarranted LVM errors
- Since pve-manager was bumped in the meantime, run the helper script
when upgrading from a version < 9.0.6 instead of < 9.0.5
- Additionally require that the script runs when upgrading from a
version >= 9.0~~
- In other words, the script is only called in postinst when upgrading
from versions where 9.0~~ <= version < 9.0.6
Additional Notes
----------------
The changes to the helper script and when it is executed are made to
reduce any potential side-effects that calling to LVM might have. If
calling `lvs` and `lvchange --setautoactivation y` fails, then
something's *really* wrong with the device anyway--and in that case, we
also dump the captured stderr.
Furthermore, if a node wasn't rebooted since new OSDs with DB/WAL were
set up, enabling autoactivation is all that needs to be done.
Even if the user rebooted the node before updating it, the update should
enable autoactivation for the OSD LVs--after that, another reboot is
sufficient to bring the OSDs back up.
Alternatively, if one wants to avoid a reboot for whatever reason, the
affected OSDs can be brought up again as follows:
1. Check for affected OSD LVs:
# lvs --options lv_name,vg_name,autoactivation,active
The names of the affected LVs should begin with "osd-wal" or "osd-db"
followed by a UUID4.
Example output:
# lvs --options lv_name,vg_name,autoactivation,active
LV VG
AutoAct Active
osd-db-2947e348-fe1b-4c38-b9d0-d24f3b8de70f
ceph-1dce1129-a411-4a14-8508-edcc8626c594
osd-wal-cc31c0cf-2b40-4ea6-afc7-eea8b767f7f5
ceph-31a0a43c-990a-40dc-9027-6412b0f6673c
osd-wal-20db00cf-b3c2-491a-bf17-4fd7c29aba6a
ceph-534964a2-b764-4ed5-a3c2-fd21aeda116a
osd-block-dd00fa96-695e-442c-99bd-ba09c2d3bd03
ceph-6bdd82ad-09bb-4c36-9382-1c302b462d7a enabled active
osd-db-42648fa6-eb1f-43db-91d3-16ef6605c62b
ceph-778869db-311b-475c-ba40-a61e531cc127
osd-block-3a178277-9ccc-488e-bfc5-4089a347195c
ceph-e7e5ac3e-56d4-4fc2-b853-7ed2465d6a69 enabled active
data pve
enabled active
root pve
enabled active
swap pve
enabled active
2. For OSD LVs for which the "AutoAct" column is empty, run the
following command:
# lvchange --setautoactivation y <vg_name>/<lv_name>
3. For OSD LVs which aren't active, run the following command:
# lvchange --activate y <vg_name>/<lv_name>
4. Double-check that autoactivation is set and that the affected LVs are
activated:
# lvs --options lv_name,vg_name,autoactivation,active
Example output:
# lvs --options lv_name,vg_name,autoactivation,active
LV VG
AutoAct Active
osd-db-2947e348-fe1b-4c38-b9d0-d24f3b8de70f
ceph-1dce1129-a411-4a14-8508-edcc8626c594 enabled active
osd-wal-cc31c0cf-2b40-4ea6-afc7-eea8b767f7f5
ceph-31a0a43c-990a-40dc-9027-6412b0f6673c enabled active
osd-wal-20db00cf-b3c2-491a-bf17-4fd7c29aba6a
ceph-534964a2-b764-4ed5-a3c2-fd21aeda116a enabled active
osd-block-dd00fa96-695e-442c-99bd-ba09c2d3bd03
ceph-6bdd82ad-09bb-4c36-9382-1c302b462d7a enabled active
osd-db-42648fa6-eb1f-43db-91d3-16ef6605c62b
ceph-778869db-311b-475c-ba40-a61e531cc127 enabled active
osd-block-3a178277-9ccc-488e-bfc5-4089a347195c
ceph-e7e5ac3e-56d4-4fc2-b853-7ed2465d6a69 enabled active
data pve
enabled active
root pve
enabled active
swap pve
enabled active
5. Finally, bring up all the OSDs again:
# ceph-volume lvm activate --all
Previous Versions
-----------------
v1:
https://lore.proxmox.com/pve-devel/20250812164631.428424-1-m.carr...@proxmox.com/T/
Summary of Changes
------------------
Max R. Carrara (2):
fix #6652: ceph: osd: enable autoactivation for OSD LVs on creation
fix #6652: d/postinst: enable autoactivation for Ceph OSD LVs
PVE/API2/Ceph/OSD.pm | 27 +++-
bin/Makefile | 3 +-
bin/pve-osd-lvm-enable-autoactivation | 176 ++++++++++++++++++++++++++
debian/postinst | 16 +++
4 files changed, 219 insertions(+), 3 deletions(-)
create mode 100644 bin/pve-osd-lvm-enable-autoactivation
--
2.47.2
_______________________________________________
pve-devel mailing list
pve-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
