[lustre-discuss] SATTR in Lustre Changelog

LEIBOVICI Thomas thomas.leibovici at cea.fr
Thu Mar 24 05:59:23 PDT 2016

On 03/23/16 18:02, Qing Zheng wrote:
> Hi-
> We are monitoring Lustre 2.7.0 using the Changelog
> and have a question on SATTR op families
> We noticed Changelog had defined SATTR, MTIME, CTIME, and ATIME log 
> types.
> We also kind of figured
> SATTR with flag 0x10 means CHMOD
> SATTR with flag 0x20 means CHOWN
> SATTR with flag 0x40 means CHGRP
> We'd like to know if it is possible for Lustre to produce SATTR 
> records with mtime/ctime/atime updates?
> And does Lustre ever produce CTIME or ATIME logs,
> since so far we don't see those logs in the Changelog even we touch a 
> file manually.

By default ATIME is not in the changelog mask. To add it:
echo "+ATIME" > /proc/fs/lustre/mdd/lustre-MDT0000/changelog_mask

AFAIK, there is a kind of priority and batching between CTIME, MTIME and 
ATIME records.
For example, if there is already a record for a given fid in the 
changelog, next ATIME/MTIME/CTIME records are not reported - until the 
first MTIME record is cleared.

# create the file
 > touch /mnt/lustre/X
# change file mtime
 > touch /mnt/lustre/X
 > lfs changelog lustre

# mtime change is not reported
327301 01CREAT 12:45:54.769600767 2016.03.24 0x0 
t=[0x200000402:0xd9fe:0x0] j=touch.0 p=[0x200000007:0x1:0x0] X

 > lfs changelog_clear lustre cl1 0
# change mtime
 > touch /mnt/lustre/X
# change atime
 > cat /mnt/lustre/X
 > lfs changelog lustre
# mtime change is reported, but not atime
327302 17MTIME 12:47:48.653207857 2016.03.24 0x7 
t=[0x200000402:0xd9fe:0x0] j=touch.0

 > cat /mnt/lustre/X
 > lfs changelog lustre
# atime change is reported
327303 19ATIME 12:49:23.43229006 2016.03.24 0x1 
t=[0x200000402:0xd9fe:0x0] j=cat.0

CTIME appears when entry ctime changes but not mtime or atime, and when 
the operation is not covered by a SATTR.
I couldn't tell exactly when this occurs, but I can see them on a 
production filesystem.


> Changelog is a great tool and we are just trying to know its behavior 
> better.
> Cheers,
> -- Qing Zheng
> _______________________________________________
> lustre-discuss mailing list
> lustre-discuss at lists.lustre.org
> http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org

More information about the lustre-discuss mailing list