[linux-lvm] Stability of pvresize

[linux-lvm] Stability of pvresize

[Louis Guillaume]

Hi,

I've seen a couple messages recently about pvresize. Since I'm about to
implement a system that might depend on this functionality, I'd like to
know whether it works well!!!

Here's the scenario:

  . Using an San.
  . A meta-LUN is created consisting of 2 `real' LUNs.
  . Meta-LUN is incorporated as a PV on the Linux Host.
  . We add a third LUN to the meta-LUN on the SAN.

This is where pvresize is supposed to come in. Once the PV is resized,
is the VG automatically aware of the new PEs? Are there any pitfalls in
this scheme?

The alternative is to not use meta-LUNs and just add a new PV as needed
and vgextend. But this puts more work on the server's CPU. Opinions?

Louis

Re: [linux-lvm] Stability of pvresize

[Zac Slade]

On Saturday 02 December 2006 11:18, Louis Guillaume wrote:
> Hi,
>
> I've seen a couple messages recently about pvresize. Since I'm about to
> implement a system that might depend on this functionality, I'd like to
> know whether it works well!!!
Works pretty damn well here.  I've used this functionality a lot and I'm 
really happy with how it's turned out.

> Here's the scenario:
>
>   . Using an San.
>   . A meta-LUN is created consisting of 2 `real' LUNs.
>   . Meta-LUN is incorporated as a PV on the Linux Host.
>   . We add a third LUN to the meta-LUN on the SAN.
>
> This is where pvresize is supposed to come in. Once the PV is resized,
> is the VG automatically aware of the new PEs? Are there any pitfalls in
> this scheme?
Perhaps if you are shrinking there might be a pitfall in making sure you are 
using a filesystem that supports shrinking as well.  However, if you are just 
growing the size there shouldn't really be any drawbacks here.  It does just 
what it advertises, gets the new size of the PV and makes adjustments to the 
PVDA and the VGDA to make the new space available to the system.

> The alternative is to not use meta-LUNs and just add a new PV as needed
> and vgextend. But this puts more work on the server's CPU. Opinions?
This is a more traditional tried and true method that will work, but you are 
correct, there will be a tradeoff in CPU usage for LE lookups during access 
across PVs.
-- 
Zac Slade
krakrjak@volumehost.net
ICQ:1415282 YM:krakrjak AIM:ttyp99