On Mon, 15 Feb 1999, Jay Soffian wrote:
We've been having a problem with Solaris NFS clients not seeing file updates occuring on our NetApps for a matter of minutes. This appears to be more than just an attribute caching issue. All our Suns are running 2.5.1 and a recently patched (Generic_103640-24). We are using NFS v3 over udp. This behavior is difficult to reproduce reliably. I managed to catch a particularly egregious example one time. A file had been updated from one of our clients, and another client didn't see the update. Out of curiousty, I mounted the NetApp partition containing the file at a differnt mountpoint on the same client that wasn't seeing the update. The client saw two copies of the same file.
Here is the _same_ file on the NetApp from two different mount points on one client:
root@presidio:~/ # ls -l {/studios,/mnt}/atlanta/accessatlanta/docs/pacman/includes/partners/rotation/namhoward.html -rw-r--r-- 1 20006 accatl 722 Dec 1 14:45 /mnt/atlanta/accessatlanta/docs/pacman/includes/partners/rotation/namhoward.html -rw-r--r-- 1 20006 accatl 49 Nov 12 10:22 /studios/atlanta/accessatlanta/docs/pacman/includes/partners/rotation/namhoward.html
I saw a very similar example here not too long ago between two Sun boxes, both running Solaris 2.5.1, the client running Generic_103640-24 and the server running Generic_103640-12. 'Cept, the files appeared the same (timestamps and md5 checksums) when they should have been different. The existing automounter mount was specious; a new mount by hand showed the true attributes and data; a umount of the automounter mount and then allowing the automounter to remount it showed truth, also.
What is that Sherlock Holmes saying....when you've eliminated all possible explanations, the one that remains must be the answer? Who would have ever thought the filesystem cache on the Solaris box would be wrong? Terribly frightening.
I guess what I'm saying is that I would point towards the client Sun box having the problem, not the filer. I forced a panic on our client and got a core dump, in case I ever have time to call Sun to report the problem.... Never saw it before; haven't seen it since.
Couldn't find anything at Sunsolve that seemed to be related to this problem.
Same here.
Until next time...
Todd C. Merrill The Mathworks, Inc. 508-647-7792 24 Prime Park Way, Natick, MA 01760-1500 508-647-7012 FAX tmerrill@mathworks.com http://www.mathworks.com ---