Re: RAC Failover testing

Greetings,

I apologize of top posting.

Anurag, I feel your testing is not correct , Looks like you are doing a "shutdown immediate"

You should be doing a shutdown abort or still better if you kill -9 either smon process
or pull the power for the box.

shutdown immediate is sort of gracefull termination, so you cannot expect that conneciton to failover.

HTH. Regards & Thank
BN

On 8/27/06, Anand Rao <panandrao_at_gmail.com> wrote:
>
> Hi Anurag,
>
> Which exact version of 9i is it? i have had nightmares with various
> versions for a simple TAF test with no load balancing, broken connection,
> etc.
>
> TAF should and will work when configured properly on 9.2 and above. it
> worked in 9i R1 also with patches. there is a good note on Metalink for
> configuring TAF, if you haven't seen it, please do read it.
>
> you have to have remote_listener parameter set correctly. the
> configuration has to be correct on both client and server.
>
> it shouldn't matter if it is shutdown immediate/abort. from a user's
> perspective its all the same. but, it works fine with shutdown abort but not
> shutdown immediate.
>
> you say that you connect to the "first node". well, did you use MYDB as
> the connect string or did you use the individual instance connect string?
>
> cheers
> anand
>
>
>
>
>
> On 27/08/06, Andrey Kriushin <Andrey.Kriushin_at_rdtex.ru> wrote:
> >
> > Hi,
> >
> > Couple of comments.
> >
> > 1. (ENABLE=BROKEN) is used to request KeepAlive feature for TCP/IP
> > socket. It might be useful if Oracle's standard DCD (Dead Connection
> > Detection) is not working properly. I've used that option several times
> > up to 9i inclusive together with OS KeepAlive settings (decreasing the
> > check interval from usual 2 hour default). It actually helped shadow
> > process to terminate when the client loses connection (well known
> > session killing problem). I've even written an article some 6 years ago.
> > Anyway, this has nothing in common with the ORA-01089, as your client
> > had successfully recognized lost connection and then occasionally tried
> > to reconnect to the same server. Which is legal as long as you provided
> > two addresses and allow LOAD_BALANCE. Well, one may expect more
> > intellectual behaviour from TAF (not try the address just faulted), but
> > that is another question.
> >
> > 2. Some warning about failed over sessions, especially when TYPE=SELECT.
> >
> > Don't forget, that specific session setting won't be restored after
> > successful failover. For example, if you'd change NLS_DATE_FORMAT from
> > instance default value via ALTER SESSION before your SELECT with DATE
> > datatype and managed to fetch some rows before the failure, than you'll
> > be surprised with the rest of results after failover. :-)
> >
> > I would not recommend to use "out of the box" TAF if you need really
> > "available" application. Take this as a demo of possibilities and go for
> > http://download-west.oracle.com/docs/cd/B10501_01/appdev.920/a96584/oci09adv.htm#449793
> >
> > There is similar chapter in JDBC documentation.
> >
> > HTH
> >
> > Andrey
> >
> > Anurag Verma wrote:
> > >
> > > Why this is happening, as I am using (ENABLE=BROKEN) and (TYPE=SELECT)
> > > options???
> > >
> >
> > --
> > http://www.freelists.org/webpage/oracle-l
> >
> >
> >
>
>
> --
> All I need to make a comedy is a park, a policeman and a pretty girl -
> Charlie Chaplin
>

-- 
Regards & Thanks
BN

--
http://www.freelists.org/webpage/oracle-l