RE: DataGuard

From: Boris Dali <boris_dali_at_yahoo.ca>
Date: Fri, 19 Jan 2007 09:17:38 -0500 (EST)
Message-ID: <996053.27612.qm@web32407.mail.mud.yahoo.com>

Laimutis,

Please see some comments in line

Laimutis Nedzinskas <Laimutis.Nedzinskas_at_landsbanki.is> wrote:

> >From: Boris Dali [mailto:boris_dali_at_yahoo.ca]
> >'Split-brain' scenario of having more than one
> database accepting connections in a broker
> configuration is prevented by the rule that two
> out of three participants in the broker config
> should agree before failover is attempted

----------------------------------------------------->

I was asking about a different scenario:
>
> Assume Observer misses network connection to the
> Primary for whatever reason. Will it assume the
> Primary dead? It should I believe. This is
> the whole purpose of it. What then prevents client A
> to connect to the Standby (which is opened
> as the Second Primary now) and client B to the
> Primary(the First Primary)?

Please consider the following two points:

1/ At least in our case observer communicates to the primary using the same network path as the users. Observer doesn't reach primary, users don't reach primary. It's that simple.

2/ 'Rule of two' I mentioned earlier still applies. Not only observer has to loose the connection to primary (one network path), but also the target standby (another/internal network path). Only if both agree FSFO is attempted and users get a chance to connect to the new primary

> The question is: how Observer is able to detect that
> standby is actually synchronized with the primary?
> What standby needs to know is that primary was not
> operating in the downgraded Maximum Performance mode
> before its death. Is this information available on
> the standby DB?

I believe the target standby needs to know that all transactions committed on primary have been applied on this standby. Up to the last committed SCN. And yes this information is readily available at the standby. I haven't traced the observer, but it would stand for a reason to think that it simply queries v$database.fs_failover_status, which should return the value of "SYNCHRONIZED" for the FSFO to be considered

Thanks,
Boris Dali.

Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com

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

Received on Fri Jan 19 2007 - 08:17:38 CST