[Lustre-discuss] e2scan wrong file list mtime/ctime

Miguel Molowny Lopez miguel.molowny at caspur.it
Wed Dec 16 05:28:55 PST 2009

Hi to everyone,
we are running lustre on our storage cluster based on IB.

We are building a "bulldozer" for a scratch area and we have found an 
odd behaviour in the e2scan utility (maybe a bug or a ... "defined 
behaviour" we don't know about).

We have reproduced this behavior several time and we are having these 
- we run e2scan on the MDT to have the files newer than 3 days ago:
   command line used was like:
     * e2scan -l -N '2009-12-13' /dev/mapper/mdt
     * e2scan -l -n /tmp/file.timestamp /dev/mapper/mdt
       where file.timestamp has been created with the date of
       3 days ago using touch
   The result is sorted and saved in a file called: 3daysago.list
- we then run another e2scan looking for files newer than 2 days ago
   using the same command line and the result is sorted and saved in
   a file called: 2daysago.list

We are looking for the differences between the first and second list to 
find files older than two days and newer than three days (that is: a
list of one day).

The problem is that there are files whose last modification time is 
before the date at which we run the two e2scan commands (they were not 
modified during the run of e2scan) _BUT_ they appear only in the list 
3daysago.list. Here is an example:

- first run of e2scan at: Dec 15 16:20 (produce list 3daysago.list)
- second run of e2scan at: Dec 15: 16:50 (produce list 2daysago.list)
- file: /scratch/foobar.txt has mtime=ctime = Dec 14 03:19
   This file is still on filesystem after the run of the two e2scan cmd
well, this file appears only in the 3daysago.list. And this happen for
a _LONG_ list of files.

For us this e2scan behavior is not trustable and we cannot use its 
results either for our bulldozer or for backup purposes.

Can anyone confirm this behaviour or help us to understand it?

Miguel Molowny Lopez
CASPUR Inter-Univ. Computing Consortium
Tel: (0039) 06/44486404
E-mail: Miguel.Molowny at caspur.it

More information about the lustre-discuss mailing list