xfs
[Top] [All Lists]

Re: Is xfsdump operation atomic?

To: xian@xxxxxxxxxxx
Subject: Re: Is xfsdump operation atomic?
From: Michael Locher <locher@xxxxxxxxxxxx>
Date: Wed, 24 Nov 2004 20:18:51 +0100
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <41A4D658.5050609@tippett.com>
References: <38A5E378-3DF4-11D9-9800-000A95BCCB96@iam.unibe.ch> <F4C07791-3DF5-11D9-9800-000A95BCCB96@iam.unibe.ch> <41A4D658.5050609@tippett.com>
Sender: linux-xfs-bounce@xxxxxxxxxxx
On Wed, 2004-11-24 at 10:43 -0800, Christian Rice wrote:
> This is why snapshots were invented.

Christian, I am well aware that this problem can be solved properly with
snapshots. My point is, that the exact behavior of xfsdump is
"underdocumented". I surprised a few colleagues when I told them that
xfsdump was not atomic... 

the following information should IMHO be included into the man pages
and/or the FAQ:

* What happens to files created/deleted (ie. directory modifications)
during a dump. (-> thanks to Ivan this has been resolved)

* What happens to existing files if they are modified (or their metadata
eg. atime) is updated *exactly* when xfsdump is working on that file?
(-> this is not yet cleared up. Maybe xfsdump could lock the file during
dump?)

I would appreciate it a lot if a developer could clarify this.

Another question: Can a frozen filesystem be dumped?

Michael


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