Re: [PVE-User] Fwd: [ceph-users] [IMPORTANT NOTICE] Potential data corruption in Pacific

["Alain péan" <alain.pean@c2n.upsaclay.fr>]

Le 28/10/2021 à 18:19, Eneko Lacunza via pve-user a écrit :
> Hi,
>
> Regarding this important issue reported for Ceph (just 
> cross-forwarding for those Ceph users not in ceph-user list):
>
> 1. Proxmox upgrade wiki page 
> https://pve.proxmox.com/wiki/Ceph_Octopus_to_Pacific marks as optional 
> disabling the conversion with "ceph config set osd 
> bluestore_fsck_quick_fix_on_mount false", and urges to convert ASAP. 
> It is now advised not to perform the conversion until a fix is released.
>
> 2. We have upgraded 2 clusters, but both have 
> bluestore_fsck_quick_fix_on_mount set to false. No command setting 
> that flag in our server's history, so I don't think we set that (I 
> wasn't the admin performin the upgrade, I'll be able to check with her 
> tomorrow)
>
> 3. Is it possible that for ceph 16.2.6 
> bluestore_fsck_quick_fix_on_mount default value is false?
>
> I expect an emergency fix released soon but I guess checking won't 
> hurt anyone ;)

Hi Eneko,

Thanks for the warning. I saw the thread on ceph-users mailing list. 
Personnally, I upgraded my clusters from PVE 6.4 to 7.0, but not yet 
from Nautilus to Pacific. I'll wait a bit for this.

Also, I have one cluster (the one in production) that is still with 
filestore OSDs. I was wondering, considering ' 
bluestore_fsck_quick_fix_on_mount", if it was mandatory to have 
bluestore OSDs on Pacific.

Personnally, when I see that you have to manually destroy your OSDs one 
after the other, then recreate it as bluestore, I am a bit reluctant to 
do the upgrade...

Let's see !

Alain

-- 
Administrateur Système/Réseau
C2N (ex LPN) Centre de Nanosciences et Nanotechnologies (UMR 9001)
10 Boulevard Thomas Gobert (ex Avenue de la Vauve), 91120 Palaiseau
Tel : 01-70-27-06-88