lvm2 weirdness in Fedora 40

lvm2 weirdness in Fedora 40

[christophe.ochal]

I'm at present fighting with LVM2, for some weird reason I can't get my
lvm volume
to be activated & mounted at boot resulting in me having to do
"vgchange -ay" at boot (after I'm dropped in a shell & prompted for my
root password,to make debugging even more troublesome I can only mess
with this over the weekends, i've included the lvmdump to this mail,
any more help would be very welcome, as i'm at a loss of how to
proceed. this might also have relevant information.

https://bugzilla.redhat.com/show_bug.cgi?id=2338735

Re: lvm2 weirdness in Fedora 40

[Roger Heflin]

The first thing I would do is change the mount line to add ",nofail"
on the options so that a failure does not drop you to single user
mode.

Then you can boot the system up and with the system on the network
figure out the state of things.

In the old days I have seen lvm2-lvmetad break systems on boot up in
bizarre ways.

I disabled it on my systems(and the 1000's of enterprise machines I
used to support) because it caused random PV's to not be found
sometimes.   Typically if something causes a pv to not get found it
will be repeatable on that given system(likely some timing problem).

The only useful thing it does is it speeds up scans when a disk is
spun down and/or when you have 1000's of disks.  But it does not speed
anything up that much unless you have a huge number of disks that are
spun down.  On large san systems the testing I did said without it it
would take 2-3 seconds to scan 1000's of disks (worth the wait given
the random failures that caused havoc), verses immediate.

And tiny changes in lvm/udev rules have changed it from working to broken.

On Sun, Feb 16, 2025 at 9:11 AM <christophe.ochal@gmail.com> wrote:
>
> I'm at present fighting with LVM2, for some weird reason I can't get my
> lvm volume
> to be activated & mounted at boot resulting in me having to do
> "vgchange -ay" at boot (after I'm dropped in a shell & prompted for my
> root password,to make debugging even more troublesome I can only mess
> with this over the weekends, i've included the lvmdump to this mail,
> any more help would be very welcome, as i'm at a loss of how to
> proceed. this might also have relevant information.
>
> https://bugzilla.redhat.com/show_bug.cgi?id=2338735
>
>
>
>

Re: lvm2 weirdness in Fedora 40

[christophe.ochal]

On Sun, 2025-02-16 at 15:45 -0600, Roger Heflin wrote:
> The first thing I would do is change the mount line to add ",nofail"
> on the options so that a failure does not drop you to single user
> mode.
> 
> Then you can boot the system up and with the system on the network
> figure out the state of things.

I wasn't aware of this option, but I'm not sure if this is any better,
right now I can just run vgchange -ay end exit to resume the boot
process, to end up in the Gnome envronment, if add nofail to /home i
stil end up in an unussable state because gnome can't load my user's
files

> In the old days I have seen lvm2-lvmetad break systems on boot up in
> bizarre ways.

I'm not sure that fedora uses lvm2-lvmetad, and google isn't helping
me,  any hits i find are for red hat 9

I wonder if this is relevant:

From lvm.conf:

 # Configuration option devices/scan_lvs.
        # Allow LVM LVs to be used as PVs. When enabled, LVM commands
will
        # scan active LVs to look for other PVs. Caution is required to
        # avoid using PVs that belong to guest images stored on LVs.
        # When enabled, the LVs scanned should be restricted using the
        # devices file or the filter. This option does not enable
autoactivation
        # of layered VGs, which requires editing LVM udev rules (see
LVM_PVSCAN_ON_LVS.)
        # This configuration option has an automatic default value.
        # scan_lvs = 1

I had no luck on googling LVM_PVSCAN_ON_LVS


> I disabled it on my systems(and the 1000's of enterprise machines I
> used to support) because it caused random PV's to not be found
> sometimes.   Typically if something causes a pv to not get found it
> will be repeatable on that given system(likely some timing problem).
> 
> The only useful thing it does is it speeds up scans when a disk is
> spun down and/or when you have 1000's of disks.  But it does not
> speed
> anything up that much unless you have a huge number of disks that are
> spun down.  On large san systems the testing I did said without it it
> would take 2-3 seconds to scan 1000's of disks (worth the wait given
> the random failures that caused havoc), verses immediate.
> 
> And tiny changes in lvm/udev rules have changed it from working to
> broken.
> 
> On Sun, Feb 16, 2025 at 9:11 AM <christophe.ochal@gmail.com> wrote:
> > 
> > I'm at present fighting with LVM2, for some weird reason I can't
> > get my
> > lvm volume
> > to be activated & mounted at boot resulting in me having to do
> > "vgchange -ay" at boot (after I'm dropped in a shell & prompted for
> > my
> > root password,to make debugging even more troublesome I can only
> > mess
> > with this over the weekends, i've included the lvmdump to this
> > mail,
> > any more help would be very welcome, as i'm at a loss of how to
> > proceed. this might also have relevant information.
> > 
> > https://bugzilla.redhat.com/show_bug.cgi?id=2338735
> > 
> > 
> > 
> > 

Re: lvm2 weirdness in Fedora 40

[Roger Heflin]

On Sat, Feb 22, 2025 at 10:05 AM <christophe.ochal@gmail.com> wrote:
>
> On Sun, 2025-02-16 at 15:45 -0600, Roger Heflin wrote:
> > The first thing I would do is change the mount line to add ",nofail"
> > on the options so that a failure does not drop you to single user
> > mode.
> >
> > Then you can boot the system up and with the system on the network
> > figure out the state of things.
>
> I wasn't aware of this option, but I'm not sure if this is any better,
> right now I can just run vgchange -ay end exit to resume the boot
> process, to end up in the Gnome envronment, if add nofail to /home i
> stil end up in an unussable state because gnome can't load my user's
> files
>
> > In the old days I have seen lvm2-lvmetad break systems on boot up in
> > bizarre ways.
>
> I'm not sure that fedora uses lvm2-lvmetad, and google isn't helping
> me,  any hits i find are for red hat 9
>
> I wonder if this is relevant:
>
> From lvm.conf:
>
>  # Configuration option devices/scan_lvs.
>         # Allow LVM LVs to be used as PVs. When enabled, LVM commands
> will
>         # scan active LVs to look for other PVs. Caution is required to
>         # avoid using PVs that belong to guest images stored on LVs.
>         # When enabled, the LVs scanned should be restricted using the
>         # devices file or the filter. This option does not enable
> autoactivation
>         # of layered VGs, which requires editing LVM udev rules (see
> LVM_PVSCAN_ON_LVS.)
>         # This configuration option has an automatic default value.
>         # scan_lvs = 1
>
> I had no luck on googling LVM_PVSCAN_ON_LVS
>
>

That is only if you put a pv on top of a vg.

Fedora may have finally got rid of lvmetad so it may not be the issue.

What does cat /proc/cmdline look like?

If /home is listed as mounting early but is not explicitly in cmdline
(either a list of LV(rd.lvm.lv=) or VG(rd.lvm.vg) to turn on at boot)
it will be missing.  And if you configured it after initial install
and/or changed the name of the LV then it won't get activated early
and will fail early.   I always use rd.lvm.vg to active everything in
the boot vg at startup.

you might try a "systemd-analyze blame" and see what it dumps for timers.

The typical disk missing timeout is like 60-90 seconds were the boot
should pause(and fail) before it gives you a emergency mode prompt.

Re: lvm2 weirdness in Fedora 40

[christophe.ochal]

On Sat, 2025-02-22 at 10:55 -0600, Roger Heflin wrote:
> On Sat, Feb 22, 2025 at 10:05 AM <christophe.ochal@gmail.com> wrote:
> > 
> > On Sun, 2025-02-16 at 15:45 -0600, Roger Heflin wrote:
> > > The first thing I would do is change the mount line to add
> > > ",nofail"
> > > on the options so that a failure does not drop you to single user
> > > mode.
> > > 
> > > Then you can boot the system up and with the system on the
> > > network
> > > figure out the state of things.
> > 
> > I wasn't aware of this option, but I'm not sure if this is any
> > better,
> > right now I can just run vgchange -ay end exit to resume the boot
> > process, to end up in the Gnome envronment, if add nofail to /home
> > i
> > stil end up in an unussable state because gnome can't load my
> > user's
> > files
> > 
> > > In the old days I have seen lvm2-lvmetad break systems on boot up
> > > in
> > > bizarre ways.
> > 
> > I'm not sure that fedora uses lvm2-lvmetad, and google isn't
> > helping
> > me,  any hits i find are for red hat 9
> > 
> > I wonder if this is relevant:
> > 
> > From lvm.conf:
> > 
> >  # Configuration option devices/scan_lvs.
> >         # Allow LVM LVs to be used as PVs. When enabled, LVM
> > commands
> > will
> >         # scan active LVs to look for other PVs. Caution is
> > required to
> >         # avoid using PVs that belong to guest images stored on
> > LVs.
> >         # When enabled, the LVs scanned should be restricted using
> > the
> >         # devices file or the filter. This option does not enable
> > autoactivation
> >         # of layered VGs, which requires editing LVM udev rules
> > (see
> > LVM_PVSCAN_ON_LVS.)
> >         # This configuration option has an automatic default value.
> >         # scan_lvs = 1
> > 
> > I had no luck on googling LVM_PVSCAN_ON_LVS
> > 
> > 
> 
> That is only if you put a pv on top of a vg.

I do recall that I have 1 PV on one Volume Group (on top of 5 Spinning
Rust drives and one  PV on top of 2 ssd's that used to function as a
cache that was setup in mirrorring 

the cache has since been retired but I have no idea how to move the
data that is on the PV on that VG to get rid of that one last layer

I hope this is clear, I'm not at the workstation in question at the
moment

> Fedora may have finally got rid of lvmetad so it may not be the
> issue.
> 
> What does cat /proc/cmdline look like?
> 
> If /home is listed as mounting early but is not explicitly in cmdline
> (either a list of LV(rd.lvm.lv=) or VG(rd.lvm.vg) to turn on at boot)
> it will be missing.  And if you configured it after initial install
> and/or changed the name of the LV then it won't get activated early
> and will fail early.   I always use rd.lvm.vg to active everything in
> the boot vg at startup.
> 
> you might try a "systemd-analyze blame" and see what it dumps for
> timers.
> 
> The typical disk missing timeout is like 60-90 seconds were the boot
> should pause(and fail) before it gives you a emergency mode prompt.

lvm2 weirdness in Fedora 40

[christophe.ochal]

[-- Attachment #1: Type: text/plain, Size: 539 bytes --]

I'm at present fighting with LVM2, for some weird reason I can't get my
lvm volume
to be activated & mounted at boot resulting in me having to do
"vgchange -ay" at boot (after I'm dropped in a shell & prompted for my
root password,to make debugging even more troublesome I can only mess
with this over the weekends, i've included the lvmdump to this mail,
any more help would be very welcome, as i'm at a loss of how to
proceed. this might also have relevant information.

https://bugzilla.redhat.com/show_bug.cgi?id=2338735



[-- Attachment #2: lvmdump-fedora-2025020884445.tgz --]
[-- Type: application/x-compressed-tar, Size: 249881 bytes --]