[linux-lvm] pvscan --cache -aay $device does not activate LV with multiple "legs"

[Tom Yan]

Hi again(x2),

Never mind, I figured it out:
https://bugs.archlinux.org/task/71385

Sorry for the noise.

Regards,
Tom

On Tue, 29 Jun 2021 at 01:25, Tom Yan <tom.ty89 at gmail.com> wrote:
>
> Hi again,
>
> I just noticed that, `pvscan --cache -aay $device` does work as
> expected if I remove `/run/lvm/vgs_online/green`. I still couldn't
> figure out why it doesn't work on boot though. (I do notice that Arch
> by default uses the "legacy" way (run the pvscan commands with a udev
> rule RUN directly; but that does not explain the problem either.)
>
> So are there any circumstances that vg(s) will be populated into
> /run/lvm/vgs_online/ without all the "legs" being up? Or circumstances
> that only the non-multi-leg LVs will be activated?
>
> Regards,
> Tom
>
> On Tue, 29 Jun 2021 at 00:47, Tom Yan <tom.ty89 at gmail.com> wrote:
> >
> > Well, it doesn't really change the game:
> >
> > [root at archlinux ~]# rm /run/lvm/pvs_online/*
> > [root at archlinux ~]# lvs | grep meh
> >   meh        green rwi---r--- 512.00m
> > [root at archlinux ~]# pvscan --cache -aay 8\:2
> >   pvscan[36958] PV /dev/sda2 online, VG green incomplete (need 1).
> > [root at archlinux ~]# pvscan --cache -aay 8\:16
> >   pvscan[36959] PV /dev/sdb online, VG green is complete.
> >   pvscan[36959] VG green skip autoactivation.
> > [root at archlinux ~]# rm /run/lvm/pvs_online/*
> > [root at archlinux ~]# pvscan --cache -aay 8\:16
> >   pvscan[36964] PV /dev/sdb online, VG green incomplete (need 1).
> > [root at archlinux ~]# pvscan --cache -aay 8\:2
> >   pvscan[36965] PV /dev/sda2 online, VG green is complete.
> >   pvscan[36965] VG green skip autoactivation.
> > [root at archlinux ~]# lvs | grep meh
> >   meh        green rwi---r--- 512.00m
> >
> > Regards,
> > Tom
> >
> > On Mon, 28 Jun 2021 at 23:08, David Teigland <teigland at redhat.com> wrote:
> > >
> > > On Sun, Jun 27, 2021 at 09:01:43PM +0800, Tom Yan wrote:
> > > > [tom at archlinux ~]$ sudo pvscan --cache -aay /dev/sdb
> > > >   pvscan[2066] PV /dev/sdb online, VG green is complete.
> > > >   pvscan[2066] VG green skip autoactivation.
> > >
> > > The "skip autoactivation" means that the VG has already been activated
> > > by a prior pvscan, so it's not being activated again.  This state is kept
> > > in /run/lvm/vgs_online/<vgname>.  The files under /run need to be cleared
> > > by reboot.  The first pvscan to activate the VG creates that temp file
> > > (during startup there's often a race among multiple pvscans to activate
> > > the same VG, and the temp file ensures that only one of them does it.)
> > >
> > > > So is this some kind of bug/regression? Or is it intended for some
> > > > reason? What I expect would be, when the command is run with any of
> > > > the legs of such an LV as the device, it will check whether all legs
> > > > of it are available and if so, the LV will be activated.
> > >
> > > The files under /run/lvm/pvs_online/ record which PVs have appeared during
> > > startup; they are created by the pvscan for each device.  pvscan uses
> > > these to know when the VG is complete and the LVs can be activated.
> > >
> > > To test this out manually, remove all the files in pvs_online and
> > > vgs_online, and run pvscan --cache -aay $dev for each device in your VG.
> > > You should find that the final pvscan will activate the VG.
> > >
> > > Dave
> > >