recommendations on how to recover a corrupted, LVM-based hard drive?

Chris Murphy lists at colorremedies.com
Sat Feb 15 17:35:13 UTC 2014


On Feb 14, 2014, at 6:32 PM, Rick Stevens <ricks at alldigital.com> wrote:

> On 02/14/2014 04:08 PM, Chris Murphy issued this missive:
>> 
>> On Feb 14, 2014, at 7:41 AM, Mark Haney <mhaney at practichem.com> wrote:
>> 
>>> -----BEGIN PGP SIGNED MESSAGE-----
>>> Hash: SHA1
>>> 
>>> 
>>> 
>>> On 02/12/2014 11:39 AM, Robert P. J. Day wrote:
>>>> 
>>>> a friend asks me if there's a way to solve the following, not out
>>>> of any sense of urgency (since there are backups) but more out of a
>>>> sense of curiosity as to whether it can even be done.
>>> 
>>> I'm a little late to the thread here because of the boondoggle that is
>>> the RTP, NC area the last couple of days, but here is my $0.02.
>>> 
>>> I learned a long time ago that recovering anything from an LVM volume
>>> is more trouble that it's worth.  I have lost more data on LVM volumes
>>> than I have any other filesystem.  Unless you use RAID with LVM never
>>> use it on a personal workstation/desktop.
>>> 
>>> In fact, I'm in the process of scrapping the LVM volume based virtual
>>> machines at my office simply because they are LVM based.
>>> 
>>> I'm sure I'll be flamed for it, but the tools to recover LVM data is
>>> humorous at best, and catastrophic at worst.
>> 
>> I think the LVM fans are on devel at . They had such a cow on devel@ a bit over a year ago that they got FESCO to override the anaconda team's decision to go with plain partitions by default with the anaconda rewrite during Fedora 18. I think LVM is very cool in many ways but subjecting users to it by default for installing an operating system is irritating. It's a hostile initial user experience. People constantly have recovery or resize problems with it, and have no idea how to use the myriad commands in the exact proper order and incantation. So it's f'n annoying to get fanboys who say it must be the default for installation and POOF they're NOWHERE to be found when people need help with recoveries and so forth.
> 
> I think LVM is quite useful for people who have large numbers of
> servers and are constantly having to deal with partitions running out of
> space and such. Is it as simple as a regular partition system? No. But
> recovering a gronked standard system can be just as arcane to the
> uninitiated.

Yes it's quite useful for all of that, and even for non-server data drive context it's useful: from a pile of whole drives as PV's, you can create LV's with their own raid level. For a single bootable disk context, I find it more complicating than helpful.

> 
> I can't say the same for silly decisions such as systemctl/systemd (who
> gives a plugged nickel how long it takes to boot your system...you
> don't do it that often) or the new, oh-so-wonderful syslog
> replacements. Or not installing an MTA by default.

Oh great another no MTA by default complaint. By default it was a waste of space, and it's easy to install an MTA of your choice. It's like complaining about GIMP not being installed by default - oh wow I might have to install it *shudder*+*panic attack*.

systemd has won on merits rather than the squeaky wheel approach. And SysV scripts were making the distributions essentially proprietary. Yes they were discoverable, but you had to become really familiar with the particular ordering used by a distribution. And for no benefit. Speaking for myself, I vastly prefer systemd and journald from a user perspective. The documentation is good. The obscurity of the startup process is less. Troubleshooting is easier. And for people with hundreds or thousands of VMs, boot times are actually kind important even if they aren't for your use case.

> 
> In many respects, Fedora has become a right pain in the arse to work
> with because the people who come up with this stuff never have to ADMINISTER machines based on it. "New" doesn't necessarily mean
> "better", gang. The Edsel was new once.

This is an old canard. People are using Fedora to admin who also develop Fedora and it's a significant basis for RHEL as well, so on the fact of it the claim is totally specious. It's fine to lament the pace of changes in Fedora, and criticize replacements on merits. But making the argument about new vs old, and familiar vs unfamiliar, are the wrong effort. It's probably a rare event, but in my view it's exactly what happen to FESCO when they rolled over to noise and history arguments rather than merits of LVM by default. That's not the case at all with systemd or MTA so the analogies just don't work.


Chris Murphy



More information about the users mailing list