On Mon, 2006-02-27 at 15:11 -0500, Ryan Taylor wrote:> I am sorry for the long *involved* post, however we are at wits end and
> advice would be greatly appreciated.
>
> Background: We have three servers in this example. Two of the servers (we
> will call Server A & B) each have a 2.7T raid 5 array that is mirrored
with
> DRBD and monitored for failover with Heartbeat. NFS is running and
> exporting the array on Heartbeats virtual IP.
> "Server C" mounts the export, and is a Samba PDC. The mount
contains all
> project data which samba hands out to ~250 windows boxes. There are at
> least ~1600 files open(locked) at any given time throughout the day. It
> works, kinda. The users can work and everything is fine except we get
> somewhat random, usually 2 freezes a day. The windows boxes freeze up for
a
> few minutes then resume working like nothing happened. The servers go idle
> during this "freeze", and sometimes we get the " lock.d
can't reach server"
> then "ok" messages in dmesg, but other times not.
Look the thread "Samba daemons hang trying to lock locking.tdb", about
Jan 25 in this list.
> The post is long enough without telling you all that we have tried. (long
> list including network drivers, nfs mount options, samba locking options,
> etc..). We now have data on "Server C" and Samba hosting it off
it's local
> drives.. It's only been one day so far, but so far (knock on wood) no
> freezes. In addition samba runs much smoother on the clients end, such as
> (*right-click*properties boxes etc...) Segue to question...
I've solved the problem reducing the amount of data shared with NFS
(making it "local" to the machine I run samba).
It's a problem (I think) with linux kernel and file looking. Have a look
to the thread "kernel BUG at fs/locks.c:1932!" of linux-kernel list,
about Fri, 17 Feb 2006.
I use latest kernels from FC4. Can you try the patch to linux kernel
from "Trond Myklebust"? I cannot try it, because my server is in
production.
I will post to this list my conclusions when I finish the investigations
about this behaviour.
> Question(s): Does Samba not work on top of NFS? Is there any better way
for
> Samba to access the data on the failover pair of servers than NFS? Is it
> locking that is causing us problems? What is the correct 'Samba'
way to get
> remote data shared to multiple samba front ends? I guess my ultimate
> question and why I am posting, is because I have tried TONS of things and
am
> curious if anyone else has this particular setup working?
I have used FC3 (with latest updates) with samba 3.0.14a, and the same
configuration (NFS, etc) and all was working ok. When I changed the
server and begin to use the latest FC4 with last versions of linux
kernel and samba, all goes very very bad.
Cheers,
--
Fermin Molina Ibarz
T?cnic sistemes - ASIC
Universitat de Lleida
Tel: +34 973 702151
GPG: 0x060F857A