Hi Alexei,

> app. can just connect to both nodes and use one which response first

But in this case you have to adjust your application (ie you cannot do this for 3rd party applications). In addition not all programs can concurrently start a connect request to all nodes. Or how do you connect to all nodes concurrently?

> And it all require very careful configuration, because listener listens on 0.0.0.0 in most cases (and many DBA configure it by these way, for reliability).

Yes, during a RAC configuration you have to be careful. But after it is configured, there shouldn't be many changes afterwards. To me using 0.0.0.0 for a listener-address is a strange way of doing it, because it is also a security risk.

> You can have NIC failures, switch failure, and many other cases of failures in real world, so these VIP transitions are not very useful anyway.

Yes there are many possibilities for failures, but you usually solve it by having redundancy. A switch failure should e.g. be transparent in terms of Oracle-connections, because for this failure you should have alternate switches (alternate paths) available. If e.g. a switch fails, the VIP is not affected at all. To me the question remains: How do you handle tcp/ip-timeouts. I.e. even keepalives run in timeouts.

> The only _reliable_ method of connections is _connect to all nodes, and use keepalives in BOTH directions_. All other methods of failover or failure detections have many scenarios when they do not work at all (and it confirms by my tests).

To me connecting to all possible nodes is not a smart solution. Suppose you have 8 nodes. Would you preconnect to all of them? What are the scenarios, when VIPs do not work at all?

Thanks

Clemens

--- Originalnachricht ---
> If VIP is used just to speed up switchover (as you describe),
> than of course
> it all is correct, but it is a very strange method to have
> such result (app.
> can just connect to both nodes and use one which response
> first).
>
> And it all require very careful configuration, because listener
> listens on
> 0.0.0.0 in most cases (and many DBA configure it by these way,
> for
> reliability).
>
> PS. You can have NIC failures, switch failure, and many other
> cases of
> failures in real world, so these VIP transitions are not very
> useful anyway.
> The only _reliable_ method of connections is _connect to all
> nodes, and use
> keepalives in BOTH directions_. All other methods of failover
> or failure
> detections have many scenarios when they do not work at all
> (and it confirms
> by my tests).
>
>
>
> ----- Original Message -----
> From: <CLEMENS.BLEILE@(protected)>
> To: "Alexei_Roudnev" <Alexei_Roudnev@(protected)
> Marin-Caea"
> <silviu_marin-caea@(protected)>
> Cc: <clemens.bleile@(protected)>
> Sent: Friday, February 10, 2006 2:56 AM
> Subject: AW: Re: Re: [suse-oracle] When should a VIP address
> move to other
> node?
>
>
> >
> > Hi Alexei,
> >
> > I think there is a misunderstanding here. A service shouldn't depend on a
> failed-over VIP.
> >
> > Let's take an example of a 2-node RAC:
> >
> > A node crashes and the VIP starts on the other node. CRS activates the
> services on the other node as well, but it registers them under
> the listener
> of the local VIP (ie the VIP of the surviving node). If a
> client connects to
> the VIP of the failed node it gets an error back immediately
> (ORA-12541 no
> listener) and tries to connect to the next VIP in the address-list
> of your
> tnsnames.ora and succeeds to connect. I.e. a VIP running on a
> different node
> than it should returns an error. The important thing is that
> it returns the
> error immediately and the client can handle it immediately
> (e.g. with TAF).
> >
> > If the second node comes back then services remain on the node which
> didn't crash (if you have configured the service
> "Preferred/Available").
> When the VIP switches back you may have a short "unavailability"
> to connect
> to the service (only to connect, not already "connected"
> sessions). On my
> Laptop with vmware this was 12 seconds. In a real environment
> probably less.
> After the failed node finished to boot I relocated the services
> back. This
> was another "unavailability" of service of 1 second. I started
> a test on my
> Laptop with 1 node up and running. To bring the service on the
> "down node"
> up again resulted in an unavailability of service of 13 seconds
> to connect
> to the service.
> >
> > Could you please explain where your long wait times come from?
> >
> > My understanding is that VIPs are there only for NIC-failures (i.e. a node
> crash means also a NIC failure ;-)). If a NIC failes then the
> VIP starts on
> another node and returns errors immediately when trying to
> connect to Oracle
> using this VIP.
> >
> > Ciao
> >
> > Clemens
> >
> >
> >
>


--
To unsubscribe, email: suse-oracle-unsubscribe@(protected)
For additional commands, email: suse-oracle-help@(protected)
Please see http://www.suse.com/oracle/ before posting