failover characteristics

[Rick Garland]

Hi all:

HPUX 11.11 on rp7410 with MC/SG 11.14

Recently did a failover to another node in the cluster. The database package itself failed over fine.

Two problems were identified however.

1) The oracle listener is not configured to failover. I am told by the powers that be that this was not possible - they do not fully understand why. Anyway, it is being considered a successful failover even though the listener did not go. End result, the end-users could not connect to the database.

Any ideas why the Oracle listener is not to be configured for failover?

2) The failover to the other node placed a copy of the production node's passswd file, cron file(s), etc, onto the failover node. I have been searching to find this step. I am wanting to have the failover make copies of the files it replaces first. What happened was once I do the manual failover back to the production node, the original files are gone. I have to restore from tape.

Where can I make this modification?

[John Poff · Accepted solution]

Hi Rick,

It sounds like your listener was configured to use the hostname of the primary host, which is bad when you failover. Our DBAs have configured the listeners here so that they either use the hostname we have assigned to the package IP address or else they put two entries in the listener config file, one for each node.

For the passwd, cron, etc. it sounds like someone might have put something into the package control file to do that. I've seen clusters where the primary box would have a cron job to copy over those key files to the failover box, so that they would be available when needed. I think you either have a package control script, a cron job, or a combination of them doing that for you. Once you find it you should be able to make it keep copies of the originals.

JP

[Sundar_7]

Rick,

1) we fail over oracle package from one node to the other all the node. Never had problems in starting the listener on the other node. I must admit I dont fully understand how Oracle works :-). but just my 2 cents

2) the one only place to look for is your package startup control script.

Look for all the steps it is executing after

if [[ $1 = "start" ]] line

- Sundar

[Rick Garland]

Some digging has produced the following.

A previous sysadmin at this site has written a script that is invoked with the failover process to do the copies of the files to the failover node. However, no copies are being made of the files. I am fixing this now.

In discussions with the DBA, he is styating that the tnsnames on the end-users PCs is configured to look specifically for the listener on the prd node. Will continue researching this one.

[A. Clay Stephenson]

Get you some new powers that be. The listener can (and should) be a part of the package. I create single instance listeners for each package and they work like a charm.
The last thing you want is for a listener to be up but not responding and that is exactly what you will have if it's not part of the package.

I'm probably in the minority but I also let the Oracle binaries move with the package as well. It uses more disk but that's a trivial concern and you do have to make certain that shared libraries are kept in sync.

[Uday_S_Ankolekar]

Try Configuring relocatabel IP or Package IP for Oracle Listner instead of pointing to any perticular node in the cluster.

-USA..

[Rick Garland]

I am on my way to deciphering. Many thanks for the directions