On Fri, Jan 31, 2003 at 11:15:17AM -0700, Rick Kelly wrote:
> D'Arcy J.M. Cain said:
> 
> >Interesting.  I would sure like some other opinions on that.  The reason for 
> >going with the NetApp is to use their snapmirror feature in order to copy a 
> >live database at an instant in time.  Using something like rsync fails 
> >sometimes because the database could change enough during the sync that it 
> >can't recover enough to start.
> 
> Back in the mid-nineties, when I worked for Progress Software, their
> software would refuse to open a database that was mounted with NFS.
> The main reason was that Progress felt that they couldn't guarantee 
> the integrity of the data. As far as I know, Oracle, Informix, and
> Sybase had similar restrictions at that time.
> 
> Has NFS changed so drastically that it no longer has data integrity
> problems?
> 
> I've used NetApp filers for lots of different jobs, but never for 
> databases.

I think the point is why would the NetBSD NFS client wedge on this type
of access?  Whether running a database with the data on NFS (or SMB or xyz)
is "good" is not relevant.  The real question is what does the postmaster
do that causes this?

Even if there is a locking issue NetBSD shouldn't just hang the mount point.
If, for some reason, the NFS client is forced to hang, we should be printing
something informative.  e.g. "NFS client attempted unsupported locking xyz"

Whatever postmaster is doing might be a good candidate for the NFS torture
test program from Apple. :-)

-Andrew