[Samba] Re: multiple wins server entries

[Christopher R. Hertel]

If the WINS servers are not synchronized, then you won't get valid results 
if you query a second WINS server if the first one fails.  That's the 
problem with disjoint name spaces.

If they *are* synced... then who knows what garblage the broken WINS
server would be sending to the backups.  Again, there is no way of knowing 
if you are getting correct answers since the authoritative source is 
messed up.

The only other way that this would work is to have all clients use two 
separate (unsynced) WINS servers.  Each client would register with each 
server, so the servers would have the same data even though they are not 
synchronizing with one another.  The problems in this case are:

1) How would you convince all of the clients to behave that way?  They're 
   all different.
2) If one server goes goofy, how do you know which one is right?

Chris -)-----

On Fri, Jun 28, 2002 at 02:57:58PM -0400, Buck Huppmann wrote:
> remember this?
> 
> On Sun, Jul 22, 2001 at 11:23:59PM -0500, Christopher R. Hertel wrote:
> 
> > The 'wins server' parameter will take a comma separated list of
> > addresses or
> > names, but at present it will only use the first.  The code is there to
> > allow Samba to fail over to the second, etc. if connections to the first
> > start to time out (indicating that the first WINS server is down).
> > 
> > I have not finished this piece because, when the failover occurs, the
> > UNICAST_SUBNET record must be updated.  I wanted to work through that
> > with
> > Jeremy before I made it 'production'.  Perhaps at the conference.
> > 
> > Some notes on what this will and won't do:
> > 
> > - It WON'T allow you to query two WINS servers to resolve a name.  This
> >   would result in disjoint NetBIOS name spaces which can really mess
> >   things up.
> 
> FWIW
> 
> we actually would have found this useful recently, when an NT WINS
> server started to play dumb and give out negative responses for all
> NBNS queries it received--including to DOMAIN<1b> queries from our
> samba machines, which made it impossible for our users to login.
> while--true--maybe we should have the bcast mechanism configured
> as a fallback in the ``name resolve order'' i think we would mind
> less having redundant unicast queries going to our on-subnet WINS
> servers in the case of incoherently-registered or bogus name queries
> than having the redundant queries be broadcast, especially since some
> of our machines are on distant networks. this seems to be what an
> h-node NT 4 SP6 machine seems to do, anyway, and maybe what we'll
> encode into our samba installations

-- 
Samba Team -- http://www.samba.org/     -)-----   Christopher R. Hertel
jCIFS Team -- http://jcifs.samba.org/   -)-----   ubiqx development, uninq.
ubiqx Team -- http://www.ubiqx.org/     -)-----   crh at ubiqx.mn.org
OnLineBook -- http://ubiqx.org/cifs/    -)-----   crh at ubiqx.org