ext3, mmap() and mtime (was RE: Redhat 7.3 bugfix)

On Mon, 22 Jul 2002, David Wojtowicz wrote:

> NB... do not make the filesystem containing your product queue or
> destination of pqact entries an ext3 filesystem.   You will pay
> severely in performance.

We have been using a mixture of systems, some ext2 and some ext3.  There
haven't been any performance differences to speak about.  It is well worth
the hassle (mostly in upgrades, not so much runtime) to get the
journalling IMO.

However I have noticed another quirk with ext3.  Apparently when you write
changes to a mmap()'ed file, the mtime (and ctime for that matter) on the
file is not updated in ext3 as it is with ext2 (and Solaris's journalled
UFS, btw).  This caused us havok when we had a watchdog script check to
see if the mtime was updating on ldm.pq before paging us about a potential
problem.

-- 
Jeff D'Angelo
ZedX, Inc. Systems Group