boot startup sequence

[John Kittel]

I recently moved a filesystem from local device to SAN NFS nount. Filesystem is used by oracle and oracle application. Filesystem mounts on /u01. Now I have the following problems booting.

I have these in rc2.d:

S430nfs.client
S99aadbora
S99zzappl

The oracle startup script, when it runs, emits a message complaining that the dbstartup script in oracle home dir does not exist. The bit of code in the script is like this:

ORA_HOME=/u01/app/oracle/product/9.2.0
if [ ! -f $ORA_HOME/bin/dbstart ]
then
echo "Oracle startup: cannot start"
exit
fi

Following that is code that starts oracle if the script is invoked with $1 = start (and of course if the script didn't exit due to the file not existing).

Somehow, even though the script is emitting the message saying oracle cannot start, it does go on and start.

I don't understand, - 1) nfsclient starts first, so why does oracle script act like the file is not found? 2) if the file is not found and it emits the message, then it looks like it should exit, so how is oracle starting anyway? 3) since oracle is starting anyway even though the script says the file isn't there, then it must mean the file really is there, right?

Next problem is with S99zzappl. This script is a link to a startup script in /u01, where the NFS filesystem should be mounted. It never executes. No "Starting application..." message on the console, and no indication in rc.log that it tried to run. Again, since nfs client started first, and then oracle started (even though it reported the file didn't exist), /u01 should / must be mounted.

[Rita C Workman]

I generally put those startups in rc3.d, at points to make them amount the last things started. Like S998....

That ensures that the system has started other things first.

Just a thought,
Rita

[Tingli]

Shouldn't there be the script to start nfs.server for mounting nfs file system?

[John Kittel]

sure, I could play around with moving oracle and oracle application startup to be later. Maybe I will try it if no other answers or suggestions come up. I have not done so yet though becaus it really doesn't seem it should be necessary. As described already, nfs client startup is in the middle of rc2.d, oracle and application are at the very end of rc2.d, with a lot of stuff in between, so there is already significant delay. There are very few things in rc3.d. If I move oracle and application to the end of rc3.d I will only be delaying their startup by a further small amount. And, as explained, /u01 does appear to be mounted when oracle tries to start, because that's where the starup script is, and it does start, even though it complains. And then application is AFTER oracle has successfully started, so really, /u01 must be available by then , mustn't it?

nfs server should not matter, I think. My hp-ux system is not exporting the filesystem. My hp-ux system is a client mounting the filesystem from the SAN.

[Steven E. Protter]

Shalom,

Check that /etc/fstab has correct settings for the oracle files system. They may not have been changed correctly.

Test them with a few umount commands followed by a mount -a

Once this happens error free, your boot should be okay.

If Oracle is on a SAN disk the start sequence of nfsclient should not matter.

If Oracle data is sitting on a NFS mount point, then nfsclient and perhaps automounter needs to be run before Oracle.

Its really a matter of understanding what services Oracle really needs and making sure they are running before you start Oracle.

SEP