[Samba] SAMBA Cluster or .......
giedz at arise.pl
Fri Mar 9 10:45:50 GMT 2007
My question is quite related to this topic. Currently I work for one of
brokerages and prepare to set up domain for all employees. However I
must be sure that if my PDC fails my BDC will provide all connections to
roaming profiles and extra shares (to be mounted during the log in
process). But additionally I want to have only ONE storage place for
placing every shares and profiles. Lucky, I "found" one EMC disk array
to be used in this solution. My idea is connect 2 server (PDC and BDC)
over FCs to ONE storage. I don't have any extra software like "cluster
Now I have almost everything UP and running but found this "thread"
about "data consistency" where different Samba instances may cause data
Please NEVER export the same file space via different Samba
nodes, this leads to data corruption because the locks
propagated have the wrong semantics...."
Now I'm wondering if I can really do this? According to what Volker said
my solution/idea is (in a word) - BROKEN :(
Does this me that I must use some extra software between EMC storage and
Samba instances to prevent data lost? In my situation there is a chance
that PDC and BDC will server THE SAME file for different users.
So looking at this topic I'd come to think that I CAN'T go production
with such prepared PDC + BDC + EMC storage - am I right? If yes would
anyone propose any solution for such idea?
Volker Lendecke napisał(a):
> On Fri, Mar 09, 2007 at 09:02:19PM +1300, Jason Haar wrote:
>> ?? doesn't that contradict the smb.conf man page?
>> "Kernel oplocks support allows Samba oplocks to be broken
>> whenever a
>> local UNIX process or NFS operation accesses a file that
>> has oplocked. This allows complete data consistency
>> SMB/CIFS, NFS and local file access (and is a very cool
>> I read that as meaning you get complete data consistency between SMB,
>> NFS and local file access :-)
>> Ahh. I just found the posting you refer to. You're saying the above only
>> works if the Samba server is also the NFS server? Not an NFS client?
>> I'm really surprised at that
> This means that yes, we get data consistency when Samba and
> other Unix processes access the same file space.
> In this sense Samba does not count as a normal Unix process
> because it has to fulfil other locking requirements on
> behalf of its clients. It is these locking requirements that
> go beyond what Posix can provide that lead to data
> corruption when multiple Windows clients access the same
> file space via different Samba instances.
More information about the samba