[Lustre-discuss] What is the proper way to rebuild a MGS/MDT node?

[Brian J. Murrell]

On Mon, 2010-04-05 at 11:29 -0500, Jeremy Mann wrote: 
> The drive in our MGS/MDT node is failing. Previously what we have done in
> the past was to copy the contents of /lustre to a backup server, replace
> the drive and recreate our Lustre environment on the MGS/MDT node as well
> as the OSTs, then copy everything back. I'm just curious if this is the
> proper way to do it.

Well, it's the long-hand way of doing it.  :-)

> Currently, the meta server is only using 570MB of space so I'm thinking of
> doing it different this time (to speed up the process). In section 15.2 in
> the 1.6 manual, there is example of using dd to block copy one device to
> another.

Yep.  As long as the target device is at least as big as the source
device.  And if it's larger, the resulting MDT won't actually be bigger,
it will just occupy the portion of the bigger device that is the size of
the MDT (i.e. the old device)[1].

> If I choose this method, is it as simple as block dd copy to the
> new drive, reboot and mount the new device as the MGS/MDT node, or are
> there other steps involved?

Nope, you've got the right idea.  Of course you *MUST* shut the MDT (at
least, if not the whole filesystem, although that is arguably overkill)
down while you dd it.

b.

[1] I *think*, theoretically you could "grow" (resize2fs) this MDT to
take up the extra space, but this is a completely untested and
unsupported operation.  I grow vanilla ext3 filesystems all the time,
however I don't think I have ever grown an ldiskfs fileystem.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20100405/44f557e4/attachment.pgp>