FreeTDS Development Group

[Frediano Ziglio <freddyz77 AT tin.it>]

> 
> > > This is much clear. I though to use syntax line
> > > Server=myserver/myinstance
> > > For mssql2k+
> > > I never used multiple instance (even on windows..). Can you configure a
> > > DSN for no default instance and see configuration entries ?
> 
> No, I don't think so.  When setting up a DSN you either have to provide a
> port number or a 'Server Alias' and determine the port dynamicly.  If you
> are specifying a port, you will get whatever instance is there.  If you are
> determining the port dynamicly, the driver setup tries a 'Server Alias'
> (instance) -> port match against a port range (not sure what this range is)
> on the server and failing that, uses port 1433.  The default instance can be
> a MSSQL v6.5, v7, or a 2k server.  If a v6.5 or v7 server is previously
> installed on a machine, then an instance of the 2k server cannot be the
> default instance.  The default instance is the only instance that uses the
> computer name as it's instance name.  All other instances are called named
> instances and their names will not be the same as the computer name.
> 
> In a multiple instance environment, if you don't know the ports, you had
> better know the instance name and determine the port dynamicly.  If you
> don't know the instance name but you do know the port, the M$ driver setup
> will magicly glean the instance name from the server.  Cute huh?
> 

Well, I did some test. To use multiple instance just set Server DSN
entry with "servername\instancename". That's all. MS driver cache port
for future use in registry so to avoid discovery every time (it cache
TDS version, port and netlib driver); cache is in our TODO.

> > <servername> listens on a host at a port.  It's the label we assign the
> > host+port combination, for convenience.  Multiple instances are merely
> > several ports on a given host.  Give them different names.  No big deal.
> > Or am I missing something?
> 
> Well...  It is just a name in the end but it is also returned by a call to
> SQLGetInfo() and it is also used to confgure the server in the Enterprise
> Manager.  Here is some output of an ODBC command line processor:
> 
> clp> connect to mstest user smurph
> Enter current password for smurph:
> 
> Connected to DSN: mstest
>  Database Name: TEST
>  Instance Name: SPM2003
>      DBMS Name: Microsoft SQL Server
>   DBMS Version: 08.00.0194
> 
> clp>
> 
> The instance name is retreived with the call:
> 
> ctl->rc = SQLGetInfo(ctl->hcon, SQL_SERVER_NAME, buffer, 255, &outlen);
> if (CTL_SUCCEEDED(ctl)) {
>     check_con(ctl);
>     oprintf(ctl, "   Instance Name: %s\n", buffer);
> }
> 
> The output is from the windows SQL Server ODBC driver.  So, the FreeTDS ODBC
> driver needs to return the instance name ('Server Alias' as M$ Calls it) for
> this call in the FreeTDS, not just the hostname.  Where does it get it from?
> Well, for one, it is returned in the TDSMSGINFO struct upon an error or
> information from the server.  But how would one know this before an error
> occurred?  I did have some code that intentionaly caused an error in order
> to glean this information from the server at the connection time.  But I
> really didn't like having to do this.  The ODBC configuration should store
> this information otherwise the above call would not return a correct 'Sever
> Alias' until it was gleaned from the server in some fashion.  It need not
> interfere with the TDS at all in the end.  Only the odbc.ini entry.
> 

Well, mssql always return some information after connection. Our ODBC
driver catch these information and set server (this is already in CVS as
part of merge work).

> Maybe we should get some terminology straight.  I'll present it in M$'s and
> their ODBC driver's setup terms.
> 

Do not confuse MS terminology with MS ODBC DSN entry... Search for "sql
server SQLDriverConnect" for info on DSN entries.

> Computer Name:  The workstation's or server's Windows name which usually
> equates to the NetBIOS netname and might be the primary TCP/IP interface's
> hostname.
> 
> Server Name:  For the TCP/IP transport, this can be the same as the
> 'Computer Name' or another local interface's hostname or a dotted quad IP
> address.
> 
> Server Alias:  It is the SAME as the 'Computer Name' for the default
> instance always.  Any other instance will have it's own name.
> 
> Port:  port is port.
> 
> So the default instance's Server Alias is alsways the same as the Computer
> Name but not always the same as the Server Name.  Throw in the fact that the
> default instance might be a downlevel server too.  A 2k server can exist
> without a default instace meaning that all instances are named.  What I'm
> getting at is that the 'Server Alias' for any given port cannot be assumed.
> It is only known by the user and the database server.
> 
> In the M$ ODBC TCP/IP Driver setup, you can specify 'Server Alias' (Instance
> Name), 'Server Name' (Which is the hostname or IP address) and a Port.
> Which leads me to believe that the 'Server Alias' is retained in the ODBC
> configuration somewhere in the registry.  And it's not in the the ODBC.INI
> DSN entry of the registr either!  But neither is port.  Only servername.
> So some weird M$ magic is happening to get the rest of the story.
> 

Search in registry... MS store these information in another registry
location. This is very bad !!! All information should be stored with
ODBC entries, not in 2 or 3 places. If I configure port and protocol
with DSN configuration tool why this information is recorded in another
location ?? Using MS DSN entries is useful for compatibility with code
that use SQLDriverConnect because in some cases you can use the same
string (normal login and FreeTDS configured with port 1433 and correct
TDS version -- this will change in next version -- I hope) but I don't
want to reproduce bad implementations !!!

freddy77