lvm2-monitor.servcie block boot process due to bad storage device

lvm2-monitor.servcie block boot process due to bad storage device

[Winston Wen]

Hi,

If a device's IO does not respond, it will block in the path of
partition scanning. But at this point, it has already created the
part0, like /dev/sda, in devtmpfs, but the mutex of the device has been
held by the partition scanning code, so all processes attempting to
open it will enter a wait.

lvm scans and opens block devices under /dev/ during startup, so if
there are any bad devices mentioned above, lvm2-monitor.server will be
blocked, and because its type is oneshot, it will block the boot
process, even if the poor device is not using lvm.

May I ask if there is a way to avoid it? For example, can the
type of this service be changed from oneshot to simple? Or is there
any other better way?

Thanks!

-- 
Thanks,
Winston

Re: lvm2-monitor.servcie block boot process due to bad storage device

[Zdenek Kabelac]

Dne 23. 04. 24 v 2:41 Winston Wen napsal(a):
> Hi,
>
> If a device's IO does not respond, it will block in the path of
> partition scanning. But at this point, it has already created the
> part0, like /dev/sda, in devtmpfs, but the mutex of the device has been
> held by the partition scanning code, so all processes attempting to
> open it will enter a wait.
>
> lvm scans and opens block devices under /dev/ during startup, so if
> there are any bad devices mentioned above, lvm2-monitor.server will be
> blocked, and because its type is oneshot, it will block the boot
> process, even if the poor device is not using lvm.
>
> May I ask if there is a way to avoid it? For example, can the
> type of this service be changed from oneshot to simple? Or is there
> any other better way?


Hi

We are aware of this weakness and we are thinking about the solution for this.

However - for booting -  you should not be actually monitoring things in ramdisk

(lvm.conf should use  monitoring=0)  - so the boot is monitore-less.


Once system is switch to rootfs - monitoring should be enabled with

vgchange --monitor y  (if enabled in /etc/lvm/lvm.conf)


But we need to rethink some caching strategy - as the code for plain lvm2 
command is now causing some troubles within dmeventd instance.


Regards

Zdenek

Re: lvm2-monitor.servcie block boot process due to bad storage device

[David Teigland]

On Tue, Apr 23, 2024 at 08:41:01AM +0800, Winston Wen wrote:
> Hi,
> 
> If a device's IO does not respond, it will block in the path of
> partition scanning. But at this point, it has already created the
> part0, like /dev/sda, in devtmpfs, but the mutex of the device has been
> held by the partition scanning code, so all processes attempting to
> open it will enter a wait.
> 
> lvm scans and opens block devices under /dev/ during startup, so if
> there are any bad devices mentioned above, lvm2-monitor.server will be
> blocked, and because its type is oneshot, it will block the boot
> process, even if the poor device is not using lvm.
> 
> May I ask if there is a way to avoid it? For example, can the
> type of this service be changed from oneshot to simple? Or is there
> any other better way?

I suggest using the devices file, or a filter (depending on your lvm
version) to prevent lvm from scanning devices other than PVs.

Dave