Re: Tale of woe - failed patch install and attempted recovery

EdStevens wrote:
> Oracle 9.2.0.1.0 on Solaris 9
>
> Began to install patchset 9.2.0.6. Our network connection had been
> dodgey all day, and when we got to the point where the installer was
> saying to go to another session as root and run root.sh .... The server
> did a connection reset on me. No panic. Just before starting the
> patch, I had asked the SA to tar up $ORACLE_HOME. Only problem is, he
> misunderstood (my fault for not being more clear) and had tar'd
> /local/home/oracle, not the /u01/...../9.2.0 where we had $ORACLE_HOME.
>
> Still no panic. We can either restore from TSM system backup or wipe
> Oracle and re-install. I feel better about the restore but the
> re-install looks much faster. We whack /u01/...../9.2.0 and fire up
> the installer. As a sanity check we looked at the 'installed
> products' tab and found it still thought there was stuff there.
> Whacked the /var/opt/oracle files. Fired up the installer, and this
> time it reports no products installed. Good to go.
>
> Selected Enterprise edition, software only. Now, when it gets to the
> actual installation phase, it reports 'uninstalling',
> 'installing', 'finished', and much more quickly than I've
> ever seen before. Uninstalling? And when it is finished, the only
> directories in $ORACLE_HOME are:
>
> oracle_at_ncens174:cd $ORACLE_HOME
> oracle_at_ncens174:ls -l
> total 8
> drwxr-x--- 2 oracle dba 512 Oct 13 10:21 bin
> drwxrwxr-x 9 oracle dba 512 Oct 13 10:44 inventory
> -rwxr-xr-x 1 oracle dba 10 Oct 13 10:36 root.sh
> drwxr-xr-x 2 oracle dba 512 Oct 13 10:21 srvm
>
> First question: Any ideas why we got this incomplete install, and why
> the installer reported 'uninstalling' after initially reporting
> nothing was installed?
>
> Second question: If the install had worked as planned, I'd still be
> missing my password file and other contents of the dbs directory. My
> plan was to recover those from the TMS system backup. Thoughts?
>
> At this point we have gone to plan B and are recovering the entire
> $ORACLE_HOME from the TSM backup taken the day prior to attempting to
> apply the patch. The SA has raised a question about
> rdbms/audit/ora_8989.aud. Apparently some active logging that caused
> the TSM backup to get spread across many tapes. ???

Well, now I really feel like an idiot (error: ID-10T) It was the OracleInventory in $ORACLE_BASE, not $ORACLE_HOME that was causing so much mischief. And the solution could have been applied immediately after the first failure of the patch installation. When the Installer listed all of the 9.2.0.6 components as installed, and greyed out on the inventory list .. don't believe they are 'really' greyed out. Click the plus sign and a good branch opens up to select for re-installing the patch for that component. Slowly work thru the entire list and it will happily re-install everything even though it thinks it has alrady done it. I feel like a rank beginner again, but at least I've learned a lot in the process -- not just the specifics of this problem, but more general about how things are put together and not believing every thing you think you see. Received on Thu Oct 13 2005 - 20:30:46 CDT