Greg Ercolano wrote:
Jeff Yana wrote:
[posted to rush.general]
Sounds like shadowy file server stuff.. possibly old data being
cached on this one machine, either an old scene file, or a bad
cached interpretation of the directory.
This might put suspicion on the file server's configuration of
opportunistic locking.
[..]
oplocks = no
[..]
I think your suspicions are correct. Many artists accessing our Linux
Server have had trouble with file systems/path names not updating in
the RUSH submit dialog, leading to much confusion (and misdirected
blame) vis a vis RUSH.
Yeah, Rush in particular brings file system synchronization problems
to light quickly, because it pulls many machines into action all at
once.
A common scenario; a user tweaks their maya scene file, saves, then
requeues the frames in their job, and 100 machines starts accessing
that file, all within a few seconds.
Even a small hiccup in file caching will become obvious instantly.
Or worse, if the artist doesn't have a healthy suspicion of computers,
they will be chasing the shadows, assuming the fault is their own;
"I just fixed that!", and will try changing other things. Then
suddenly,
for no apparent reason at all, it all suddenly starts working when the
caches expire.
It's stuff like this that frazzles artists if it happens enough..
their hands visibly shaking at the coffee machine..
I guess this is where an investment in an expensive NetApp file server
comes back as a savings to everyone's sanity.
There has even been a whole host of problems getting Windows explorer
on some machines to accurately refresh, or not refresh at all.
By 'explorer', I imagine you mean the folder browsers in Windows; yes.
They are automatically supposed to update when the file server tells
them files have been changed or added.
Example: if a linux client creates a new directory on the file server,
the file server's /kernel/ is supposed to notify samba, and /samba/
is supposed to notify any clients that are 'watching' that directory
tree,
(eg. window's 'explorer' browser) so that they update their on-screen
directory listings automatically. (Otherwise you have to hit F4 to
poll it)
If you have a commercial file server, beat on their support, because
they should have configured all this for you with defaults consistent
with Microsoft's own, as that's what you're paying for.
In fact, it is a commerical product that WAS configured by the vendor.
Perhaps the problem lies with SAMBA's implementationn of OPSLOCK, as I
have never seen this issue with NetApp, which uses their own proprietary
CIFS protocol, which IMHO is better than either Samba or MS's product by
a long shot...
I will take a look at some of this documentation below. Probably would
not hurt to do a review.
Cheers!
If you have a custom file server /you/ set up, then look closely at
the samba docs on this stuff;
For file caching, see the 'OPLOCKS' sections (level 1 and 2)
in Speed.txt (as well as other sections in that doc, eg. the
TCP_NODELAY socket option) and the smb.conf(5) man page.
For folder browsers not updating, see 'notify timeout'
in smb.conf(5), and related flags. For kernel related docs
on directory notification (which I *think* is what samba must use),
see /usr/src/linux/Documentation/dnotify.txt.
For complete information saturation, see the Samba mailing list
archives.
|