[Top] [All Lists]

XFS + LVM Root fs questions

To: linux-xfs@xxxxxxxxxxx
Subject: XFS + LVM Root fs questions
From: Kai Leibrandt <k_leibrandt@xxxxxxxxxxx>
Date: Tue, 29 Jun 2004 20:40:57 +0200
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Mozilla Thunderbird 0.7 (Windows/20040616)
Hi all,

I was just wondering if I could get some general feelings from people with experience with xfs on top of lvm2 (pref. on top of md raid1 pv's), on the following;

So far I have been running a smallish (ca. 150Gb) mail/samba server very happily on plain xfs on md raid1. I have been doing nightly backups just xfsdumping to a seperate disk (which holds only the backups, no raid mirrors), and this has been very reliable and convenient. I am doing these xfsdumps for the / partition as well as the seperate /home partition (which holds all user data, including their maildirs).

I have recently been reading up on backup strategies and found a few references to people using lvm, who do an xfs_freeze, take a lvm snapshot, xfs_freeze -u, then mount the snapshot ro, then do an xfsdump of the ro-mounted snapshot, finally unmount and discard the snapshot.

I do realize the benefits of doing this (i.e. no changed data between taking the xfsdump's inventory and the actual dumping of the potentially now-changed data), and I am seriously considering converting my whole setup to xfs->lvm2->md raid1.

So my question is: are there any issues with this setup that I should be aware of? Would it be okay to have the / as well as the /home partitions managed by lvm or is the general consensus to leave / alone on a plain-jane xfs-> md raid1? In particular, are there any issues that might hit me in the forehead when xfs_freeze-ing / (if only for the few seconds it takes to do the lvm snapshot)?

Many thanks for any input,


PS: I am running a 2.6.6 kernel, and fairly up-to-date (though non-CVS) xfs code; xfsdump 2.2.16, xfsprogs 2.6.3, no lvm yet but planning to use lvm2.

<Prev in Thread] Current Thread [Next in Thread>
  • XFS + LVM Root fs questions, Kai Leibrandt <=