[linux-lvm] vgcreate doesn't work with lvmlockd running

[Zhiyong Ye]

Hi David,

Thanks for your reply and detailed answer. It is true that It works fine
with the latest version of lvm as you said.

Thanks again!

Zhiyong

在 5/5/22 10:07 PM, David Teigland 写道:
> On Thu, May 05, 2022 at 02:08:36PM +0800, Zhiyong Ye wrote:
>> root at n151-143-035:~# lvmlockd -V
>> lvmlockd version: 2.03.02(2) (2018-12-18)
>> root at n151-143-035:~# sanlock -V
>> sanlock 3.6.0 (built Jul 11 2018 12:44:29)
> 
>> It is mentioned in the main page under "creating the first sanlock VG"
>> that the first VG is created to get the global lock, so I create the VG
>> directly:
>>
>> root at n151-143-035:~# vgcreate --lock-type sanlock --shared test
>> /dev/mapper/mpatha1 --verbose
> 
> This is correct.
> 
>>    Enabling sanlock global lock
>>      Wiping signatures on new PV /dev/mapper/mpatha1.
>>      Set up physical volume for "/dev/mapper/mpatha1" with 2147483648
>> available sectors.
>>      Zeroing start of device /dev/mapper/mpatha1.
>>      Writing physical volume data to disk "/dev/mapper/mpatha1".
>>    Physical volume "/dev/mapper/mpatha1" successfully created.
>>      Adding physical volume '/dev/mapper/mpatha1' to volume group 'test'
>>      Archiving volume group "test" metadata (seqno 0).
>>      Creating logical volume lvmlock
>>      Creating volume group backup "/etc/lvm/backup/test" (seqno 2).
>>      Activating logical volume test/lvmlock.
>>      activation/volume_list configuration setting not defined: Checking only
>> host tags for test/lvmlock.
>>      Creating test-lvmlock
>>      Loading table for test-lvmlock (254:1).
>>      Resuming test-lvmlock (254:1).
>>    /dev/test/lvmlock: not found: device not cleared
>>    Aborting. Failed to wipe start of new LV.
> 
> This looks like a bug, and it does look familiar.  I'd guess that section
> of code is missing a call to synchronize with udev (sync_local_dev_names).
> Unfortunately your version of lvm (and sanlock) are old enough that it's
> hard to say immediately if it's fixed.  The best option would be trying
> newer versions where it's most likely been fixed.
> 
> Dave
>