On Fri, 4 Feb 2005, Roberto wrote:
Yes, this could be an interesting test: I remember (some times ago) on
this list a similar problem where a deleted files where still keep open by
a daemon process: this led to df - du to disagree...
But I've not enough info to say it's the same problem ...
The correct behaviour for a filesystem on Unix is for the kernel to hold
on to the pointer to the file while it is still being kept open by a
process even if the file has had its link count reduced to zero (ie,
The other day I accidentally deleted a tar file I was compressing (with
bzip2). This was on an xfs filesystem too. I didn't worry (no need to
pull it from backups) as bzip2 happily completed the compression. After
it had finished it told me its attempt to remove the tar file failed. I
would only have had a problem if I had killed the bzip2 process.
Getting back to the original poster, it is well known that differences
betweeen du and df output can be a result of holes in a file. I have seen
this with empty swap files and I'm not necessarily surprised if a database
file is exhibiting this behaviour.
Robert Brockway B.Sc.
Senior Technical Consultant, OpenTrend Solutions Ltd.
Phone: 416-669-3073 Email: rbrockway@xxxxxxxxxxxxx http://www.opentrend.net
OpenTrend Solutions: Reliable, secure solutions to real world problems.
Contributing Member of Software in the Public Interest (http://www.spi-inc.org)