FreeTDS Development Group

[Frediano Ziglio <freddy77 AT gmail.com>]

2016-01-07 9:31 GMT+00:00 esterniclos AT gmail.com <esterniclos AT gmail.com>:
> Hello everyone,
>
> We would like to adapt freetds to SQL Server Always On Clustering.
>
> This kind of cluster is pretty simple:
>
>
>         Every database has associated a Listener with a dns name. E.g.
>         dblistener.myorg.com
>
>         Every listener has 2 or more IPs associated:
>         nslookup dblistener.myorg.com responds something like: 10.57.3.2
>         and 10.57.4.52
>
>
>         Only one of the IP's serves the database (active-passive)
>
>         The connection has to be made directly to the listener. and use
> the IP that is working at that moment.
>
> TDS Behaviour at this point.
>
>     Right now, TDS works as long as the listener is in the first node.
> In the example
>                  dblistener.myorg.com   instance working in  10.57.3.2
>
>     If the database is moved to another instance, it returns a timeout.
> Never checks the rest of IPs
>
> Possible solution
>
>         Try to login in several times with as many objects as IP's are
> in DNS:
>
>                         tds/login.c:tds_set_host(TDSLOGIN * tds_login,
> const char *hostname)
>
>
>           This would penalize a little when the database is in the
> backup instance, as it has to receive a timeout before making the
> connection.
>
>           To speed up things, the last IP where the database was seen
> could be also stored, and only ask dns if there is a time out.
>
> Thank you for your time, and any advice is welcome.
>
>                Ester Niclós Ferreras


Now (master) we try to connect to all IPs, not just the first.
Yes, could be cached the last success somewhere but is a start.

Frediano