Emond Papegaaij
2018-07-18 12:59:49 UTC
Hi,
On a CentOS 7.5 based virtual appliance we are developing, I want to use LVM
snapshots to be able to rollback failed upgrades. Prior to the upgrade, I
create a snapshot volume with:
lvcreate -s /dev/tkhsystem01/root -l100%FREE -nroot_snap
On a failed upgrade, I merge this snapshot and reboot with:
lvconvert -y --merge /dev/tkhsystem01/root_snap
reboot
This works fine, except it causes a 90 second delay in shutdown where systemd
waits for lvm2-lvmetad.service to shtudown. This fails and the process is
killed after which shutdown continues. After boot, the snapshot is restored
correctly. Reading through some reports, I found that monitoring of the
snapshot volume might cause lvmetad to not shutdown, so I added the following
command just prior to lvconvert:
lvchange --monitor n /dev/tkhsystem01/root_snap
This fixes the 90 second timeout at shutdown, but it is unclear to me what
possible side effects it may have. To be clear, the lvchange, lvconvert and
reboot commands are executed in a script in direct succession. Does it pose
any danger to stop monitoring just before a reboot? Is this a bug in LVM2 or
CentOS or am I doing something wrong?
Best regards,
Emond Papegaaij
On a CentOS 7.5 based virtual appliance we are developing, I want to use LVM
snapshots to be able to rollback failed upgrades. Prior to the upgrade, I
create a snapshot volume with:
lvcreate -s /dev/tkhsystem01/root -l100%FREE -nroot_snap
On a failed upgrade, I merge this snapshot and reboot with:
lvconvert -y --merge /dev/tkhsystem01/root_snap
reboot
This works fine, except it causes a 90 second delay in shutdown where systemd
waits for lvm2-lvmetad.service to shtudown. This fails and the process is
killed after which shutdown continues. After boot, the snapshot is restored
correctly. Reading through some reports, I found that monitoring of the
snapshot volume might cause lvmetad to not shutdown, so I added the following
command just prior to lvconvert:
lvchange --monitor n /dev/tkhsystem01/root_snap
This fixes the 90 second timeout at shutdown, but it is unclear to me what
possible side effects it may have. To be clear, the lvchange, lvconvert and
reboot commands are executed in a script in direct succession. Does it pose
any danger to stop monitoring just before a reboot? Is this a bug in LVM2 or
CentOS or am I doing something wrong?
Best regards,
Emond Papegaaij