[Samba] static wins entries for failover shares -- do Microsoft clients get confused?

Christian Brandes christian.brandes at forschungsgruppe.de
Fri Oct 12 14:52:12 GMT 2007

On my RedHat Cluster I am trying to configure samba shares that can be
relocated or fail over from one node to anoter.

This is done by:
1. mounting the corresponding file system on the node from SAN
2. having the cluster assign the corresponding virtual ip number to that
3. I tried to run an aditional instance of samba for that share
( See posting: "security = domain -- samba adds its netbios name as
samba domain to LDAP")

As I ran into trouble with such a lot samba instances beeing safely
started and stopped on one machine, now I try another method:

I try to run only one instance of samba on each node and then work with:
"netbios alias =" and "include ="

It is not possible to add all needed netbios aliases to all nodes because
then all nodes seem to have the same names and clients get confused,
trying to resolve names and having to wait for some timeout.
I do not know what happens in detail. What I know is that sometimes it
takes a long time to establish a file transfer, but when it is
established, it has full bandwidth. So it is more some kind of latency.
This meens processes that open a lot of files can drastically slow down.

So I looked for the solution in the use of wins. I eliminated "netbios
aliases" in smb.conf. Instead I introduced "Static WINS Entries" as
stated in Samba-3-Howto like that:

"SERVER1#00" 0 66R
"SERVER1#20" 0 66R
"SERVER1#03" 0 66R
Which is the netbios name of my one of my servers with it's two physical
ip numbers on it's network interfaces.

"SHARE1#00" 0 66R
"SHARE1#20" 0 66R
"SHARE1#03" 0 66R
Which is the former netbios alias associated with the two virtual ip
numbers that are assigned to the node that actually holds the
corresponding share. The share should be accessible by: \\SHARE1\share1
Distinguished is by "include = /etc/samba/smb.conf.%i".

So that works fine. Name resolution is done by wins real quickly. For
some reason a DNS query as well is done by Microsoft Clients even when
the name is resolved by wins. And hybrid clients even do a broadcast
after they got their wins reply.

For some reason after a while Microsoft clients still get confused.
Maybee when they get a wins reply, then broadcast and get a different
one from the server. This affects even file transfer to Novell Clients,
which sometimes have this latency, too. So it must be a blocking wait
for a timeout.

( See posting: multihomed samba -- how to tell samba which ip-numbers
to respond? )

Does anyone know why Microsoft clients get confused?
Or even better how to configure a cluster with samba shares that fail
over from one node to another?

Best regards

More information about the samba mailing list