On August 13, 2025 11:40 am, Max R. Carrara wrote: > On Wed Aug 13, 2025 at 11:02 AM CEST, Fabian Grünbichler wrote: >> On August 13, 2025 10:50 am, Max R. Carrara wrote: >> > On Wed Aug 13, 2025 at 9:52 AM CEST, Fabian Grünbichler wrote: >> >> On August 12, 2025 6:46 pm, Max R. Carrara wrote: >> >> > Introduce a new helper command pve-osd-lvm-enable-autoactivation, >> >> > which gracefully tries to enable autoactivation for all logical >> >> > volumes used by Ceph OSDs while also activating any LVs that aren't >> >> > active yet. Afterwards, the helper attempts to bring all OSDs online. >> >> >> >> I think this is probably overkill - this only affects a specific non >> >> standard setup, the breakage is really obvious, and the fix is easy: >> >> either run lvchange on all those LVs, or recreate the OSDs after the fix >> >> for creation is rolled out.. >> >> >> >> i.e., the fallout from some edge cases not being handled correctly in >> >> the 200 line helper script here is probably worse than the few setups >> >> that run into the original issue that we can easily help along >> >> manually.. >> > >> > I mean, this script doesn't really do much, and the LVs themselves are >> > fetched via `ceph-volume` ... But then again, you're probably right that >> > it might just break somebody else's arcane setup somewhere. >> > >> > As an alternative, I wouldn't mind writing something for the release >> > notes' known issues section (or some other place). Assuming a standard >> > setup*, all that the user would have to do is identical to what the >> > script does, so nothing too complicated. >> >> but the known issue will be gone, except for the handful of users that >> ran into it before the fix was rolled out.. this is not something you >> noticed 5 months later? > > Are you sure, though? There might very well be setups out there that are > only rebooted every couple months or so--not everybody is as diligent > with their maintenance, unfortunately. We don't really know how common / > rare it is to set up a DB / WAL disk for OSDs.
I am not opposed to adding it to the known issues, I just think it's not a very common issue to run into in the first place. >> > (*OSDs with WAL + DB on disks / partitions without anything else inbetween) >> >> I am not worried about the script breaking things, but about it printing >> spurious errors/warnings for unaffected setups. > > Well, unaffected here would mean that all OSD LVs have autoactivation > enabled (and are also activated). Additionally, if the OSDs are already > up, `ceph-volume` doesn't do anything either. no, unaffected here means 99% of the systems. and as soon as you run commands on any system, there will be a small percentage where those commands fail in noisy fashion for whatever reason. so the tradeoff needs to be there - if the issue is big enough and the commands are not very involved, then yes, fixing in postinst like this is a good idea. for an issue that will affect almost nobody, and that requires running commands with lots of storage interaction(!), that tradeoff looks rather different.. > FWIW we could suppress LVM "No medium found" warnings in the LVM > commands the script runs, like we do in pve-storage [0]. Additionally, > we could also short-circuit and silently exit early if no changes are > necessary; e.g. if autoactivation is enabled for all OSD LVs, we can > assume that they're activated as well, and that the OSDs themselves are > therefore running, too. > > So, I would really prefer having either (an improved version of) the > script, or some documentation regarding this *somewhere*, just to not > leave any users in the dark. > > [0]: > https://git.proxmox.com/?p=pve-storage.git;a=blob;f=src/PVE/Storage/LVMPlugin.pm;h=0416c9e02a1d8255c940d2cd9f5e0111b784fe7c;hb=refs/heads/master#l21 > >> >> >> _______________________________________________ >> pve-devel mailing list >> pve-devel@lists.proxmox.com >> https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel > > > > _______________________________________________ > pve-devel mailing list > pve-devel@lists.proxmox.com > https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel > _______________________________________________ pve-devel mailing list pve-devel@lists.proxmox.com https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
