Re: Opinions on ext3 vs XFS vs reiserfs for LAMP server
On 08/23/07 04:07:46PM -0700, michael@estone.ca wrote:
> Quoting Jim Crilly <jim@why.dont.jablowme.net>:
>
> >On 08/23/07 10:03:24AM -0700, michael@estone.ca wrote:
> >>
> >>The problem of zeroing files of XFS still exists, however its not some
> >>mythical type of corruption. You'll only see it on files recently
> >>written to within seconds (say approx 60 secs) of a hard power off. If
> >>you can't risk it, or think you may have encounter the odd hard reset,
> >>ext3 might be a better choice.
> >>
> >
> >Actually it's been fixed as of 2.6.22:
> >http://oss.sgi.com/projects/xfs/faq.html#nulls
> >
> >Of course that doesn't help you if you're using sticking with the kernel
> >shipped with etch.
> >
>
> I'm not so sure its fixed.
> I just tested with a sid samba box, running 2.6.22 kernel, and XFS
> filesystem.
> Connected to it via a WinXP box and copied a word doc file to it.
> Soft rebooted the samba box to make sure the file was sync'd to hard drive.
>
> Re-connected to samba share and opened the word document, added some
> text lines to it, saved and quit Word, then yanked the power out.
> Rebooted and re-connected to the samba share again only to find the
> file full of squares.
> Ext3 would have at least retained the original contents of the file.
>
> I tested the exact same thing again but waited 60 seconds after saving
> the file, and then yanked the power out. Upon a boot up, the file was
> intact and the save worked. So you still have about a 60 second window
> of newly written files and a power loss for data corruption, unless
> the program can sync it to disk before that.
>
Well I'm only passing on what the XFS devs have said, all of my boxes are
on UPSes so I rarely saw the issue anyway. But are you sure the squares you
saw in the word doc were nulls? The FAQ page says that you can use the
xfs_bmap command to see if it has any extents allocated and if it does then
it would likely be another issue.
Jim.
Reply to: