Re: [TCLUG:16067] problems with df/du mismatch

[Chris McKinley <lamfada@lugh.net>]

That is because you did not restart the apache process.  Until you do that, 
there is a valid reference to those files in memory[1], even though there is 
no inode entry in the directory.  So du will not see it, but df will.

-Chris McKinley

[1] See any UNIX kernel book, a good one is "Design and Implementation of 4.4 
BSD"

On Wed, 12 Apr 2000, Adam Maloney wrote:

> One of my machines generated a bunch of apache logfiles and the
> partition became 100% full.  I moved the logfiles to a different machine
> and removed them.  The logfiles were 300MB each (2 of them), the
> partition is about 3Gb total.  After I removed them I did a DF, and the
> partition still showed up as 100% in use.  I removed a smaller 30MB file
> and brought the usage down to 99%.
> 
> When I do a DU from /usr, I see: 2193978	.  That tells me that DU sees
> 2,193,978 bytes in use for the whole partition.  When I do a df, I see:
> /dev/sda8            2971111 2800051    17410     99%   /usr
> 
> So df thinks that there's 2,800,051 bytes in use.
> 
> Incidentally, before I deleted the 30MB logfile (after removing the 2
> 300MBers and having df say it was still 100% full) I tried to write a
> file and got the no space left on device message.  So I assume that when
> the 17MB that's currently free gets used up by the end of the day
> tomorrow Linux is going to think the partition is full.
> 
> Do I need to unmount the partition and fsck to fix the mismatch?