Corrupted system.reg in clustered environment - where to look?

Hello Wine developers,

we try to use Wine in clustered environment with multiple 
Wine processes being started simultaneously on 8 to 80 Linux 
machines by a modified GNU make. The source code being built 
is located on an NFS mount visible by all cluster machines.

Unfortunately Wine fails after some time because of 
corrupted ~/.wine/system.reg. I guess multiple Wine processes
try to update some keys in that file, like (differing) 
processor steppings etc. and this goes wrong at some point.

Also I've noticed, that the registry is being re-saved every 
30 seconds (why?) - here in wine-0.9.32/server/registry.c:

	static const int save_period = 30000;
           /* delay between periodic saves (in ms) */

I've tried adding file locking using fcntl() (which should
work over NFS too) by modifiyng load_registry() and 
save_registry() as in the attached diff, but unfortunately
it doesn't help.

Does anybody have a hint where to look?

Why updating registry is needed at all and can I disable
it maybe? (And where would be the good spot to do that?)

Thank you
Alex

[Alexandre Julliard]

<Alexander.Farber(a)nokia.com> writes:
> I've tried adding file locking using fcntl() (which should
> work over NFS too) by modifiyng load_registry() and 
> save_registry() as in the attached diff, but unfortunately
> it doesn't help.

For the standard registry files you'd have to do it in
load_init_registry_from_file() and save_branch(). It still won't
prevent one server from overwriting changes made by others of course.
> Why updating registry is needed at all and can I disable
> it maybe? (And where would be the good spot to do that?)

If you make the WINEPREFIX directory and the registry files read-only
that should prevent the server from updating them.

-- 
Alexandre Julliard
julliard(a)winehq.org

Hi Alexandre,
> From: ext Alexandre Julliard [mailto:julliard(a)winehq.org]
>
> For the standard registry files you'd have to do it in
> load_init_registry_from_file() and save_branch(). It still won't
> prevent one server from overwriting changes made by others of course.

thanks, you are correct: the load_registry() and 
save_registry() aren't called in my case at all - I was 
modifying wrong functions. Here I've added few fprintf's:

salsim04:afarber {528} wine perl -v
load_init_registry_from_file(/disks/ouusers/afarber/.wine/system.reg,
key)
load_init_registry_from_file(/disks/ouusers/afarber/.wine/userdef.reg,
key)
load_init_registry_from_file(/disks/ouusers/afarber/.wine/user.reg, key)
save_branch(key, /disks/ouusers/afarber/.wine/system.reg)
save_branch, renaming /disks/ouusers/afarber/.wine/reg205b0000.tmp ->
/disks/ouusers/afarber/.wine/system.reg
save_branch(key, /disks/ouusers/afarber/.wine/userdef.reg)
save_branch, renaming /disks/ouusers/afarber/.wine/reg205b0000.tmp ->
/disks/ouusers/afarber/.wine/userdef.reg
save_branch(key, /disks/ouusers/afarber/.wine/user.reg)
save_branch, renaming /disks/ouusers/afarber/.wine/reg205b0000.tmp ->
/disks/ouusers/afarber/.wine/user.reg

This is perl, v5.6.1 built for MSWin32-x86-multi-thread
(with 1 registered patch, see perl -V for more detail)

Copyright 1987-2001, Larry Wall

Binary build 635 provided by ActiveState Corp.
http://www.ActiveState.com
Built 15:34:21 Feb  4 2003

Perl may be copied only under the terms of either the Artistic License
or the
GNU General Public License, which may be found in the Perl 5 source kit.

Complete documentation for Perl, including FAQ lists, should be found on
this system using `man perl' or `perldoc perl'.  If you have access to
the
Internet, point your browser at http://www.perl.com/, the Perl Home
Page.

salsim04:afarber {529} save_branch(key,
/disks/ouusers/afarber/.wine/system.reg)
save_branch, renaming /disks/ouusers/afarber/.wine/reg205b0000.tmp ->
/disks/ouusers/afarber/.wine/system.reg
save_branch(key, /disks/ouusers/afarber/.wine/userdef.reg)
save_branch(key, /disks/ouusers/afarber/.wine/user.reg)
save_branch, renaming /disks/ouusers/afarber/.wine/reg205b0000.tmp ->
/disks/ouusers/afarber/.wine/user.reg


However now I'm really lost, because I don't understand
where does the file corruption occur. In my case the
~/.wine/system.reg is a regular file stored in my NFS-
mounted home dir. That is why a tmp file is used and
later renamed. Shouldn't that operation be atomic and
if not, how could I protect it (locking a dummy file?)

Regards
Alex

PS: BTW I've noticed a bug - the variable "count" in
    the save_branch() is probably meant to be static.
    But it is not and that's why the tmp file's name 
    (the "reg205b0000.tmp" above) never changes.