[Lustre-discuss] timestamp problem in lustre?

[Andreas Dilger]

On Nov 02, 2008  17:00 +0800, David Levi Hevroni wrote:
> I encounter the following problem when trying  to compile a program on a
> lustre volume:  when I do ¨make¨, the program compiles. But then, when I
> switch to another computer and run make again, I get the ¨clock skew
> detected¨ warning, but only on the executable that was created with the
> first ¨make¨.

The timestamps in lustre are based on the client clock.  That ensures
that when running programs like "make" that you don't get clock-skew
problems between the client and server.

> When I take a look at the timestamp of the file, I see a valid time on the
> first computer, and just the day on the second. does anyone have any idea
> why such thing might happen?

Can you check "ls -l --full-time {file}" on both clients?  When just the
day is shown from "ls -l" it usually means that the file was modified so
long ago that the time is no longer relevant (depending on the output format
chosen for your locale).

> This happens when I use NTP daemon, and also when I am synchronizing the
> clocks by myself. Running ¨date¨ on both computers gives the same time
> (maybe 1 second difference).

This is very odd.

> Can it be a problem with the server? Does the date on the lustre server has
> any importance in this matter?

The server time should NOT affect the timestamps on the files, but it
isn't impossible that there is a bug in this area.  There were a few
timestamp bugs fixed recently, for the cases where a file was renamed
after it was created incorrectly changing the timestamp.

Please also check the time on the servers (both MDS and OSTs).

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.