xfs
[Top] [All Lists]

Re: re[2]: FWD: fsck in background?

To: Greg Freemyer <freemyer@xxxxxxxxxxxxxxxxx>
Subject: Re: re[2]: FWD: fsck in background?
From: Chris Wedgwood <cw@xxxxxxxx>
Date: Wed, 9 Apr 2003 16:01:53 -0700
Cc: xfs mailing list <linux-xfs@xxxxxxxxxxx>
In-reply-to: <20030409225042.IBUO25644.imf43bis.bellsouth.net@tiger2>
References: <20030409225042.IBUO25644.imf43bis.bellsouth.net@tiger2>
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Mutt/1.3.28i
On Wed, Apr 09, 2003 at 06:53:47PM -0400, Greg Freemyer wrote:

> I will add a fragmentation test.  If it is too bad I will log/e-mail
> and run xfs_fsr.

xfs_fsr is a terrible tool in a sense

people see it and think 'ah cool, lets defragment my filesystem... we
do this in windows and thats good'

by and large, xfs shouldn't need lots of defragmentation --- that
said, NFS users do see lots of fragmentation at present because
synchronous writes don't preallocate presently

> ie. this would not be useful:
>
> create readonly snapshot
> if [xfs_repair -n snapshot != 0]
>    email admin

it's pointless mostly

"xfs_repair -n" will miss things that it otherwise would detect on a
RW fs

also, you need to consider this is a very slow process... how big is
your fs anyhow?


> 2) With a read/write snapshot, something like the below may work.
>
> Weekly (or monthly)
>
> create r/w snapshot
> mount snapshot r/w (and repair any issues caused by the snapshot process)
> unmount snapshot
> if (xfs_repair != 0)
>    email admin

if the snapshot is created with xfs_freeze, then I think the log will
be in an active state you you'll want xfs_replair -L

> Updating the script seems easy enough, but if r/w snapshots are
> needed to run/test it, I doubt I will make the effort.  (i.e. I
> think you need a DM enabled kernel for that, right?)

i think you need to consider what you're trying to achieve here

there are multi-TB and probably multi-PB XFS filesystems in use and
they don't do these sorts of extreme things (nor could they when the
fs gets that large) --- it's better to address the problems if you
find them


  --cw


<Prev in Thread] Current Thread [Next in Thread>