Patching Oracle without shutting down the database

Somehow it appears to be possible to patch Oracle on AIX 5.3 without shutting down the databases - and everything works fine! - We have several AIX 5.3 servers running Oracle 9.2.0.7.0 Enterprise Edition - Partitioning Option wasn't installed - I installed partitioning option by running 9.2.0.1.0 Installer, then 10g Installer then comes with 9.2.0.7.0 Patch Set, and then Patch 5496862 using OPatch - and then realised that the database was up. - Everything seems OK, there were no errors during installation, timestamp on the file $ORACLE_HOME/bin/oracle shows that it was re-created, I can shut down and start the database, partitioning works - The experiment was repeated on the second server with 6 databases running - same result. Is it a feature of AIX? Thanks Sev

Vsevolod Afanassiev wrote: > Somehow it appears to be possible to patch Oracle on AIX 5.3 without > shutting down the > databases - and everything works fine! > - We have several AIX 5.3 servers running Oracle 9.2.0.7.0 Enterprise > Edition > - Partitioning Option wasn't installed > - I installed partitioning option by running 9.2.0.1.0 Installer, then > 10g Installer then comes with 9.2.0.7.0 Patch Set, and then Patch > 5496862 using OPatch - and then realised that the database was up. > - Everything seems OK, there were no errors during installation, > timestamp on the file $ORACLE_HOME/bin/oracle shows that it was > re-created, I can shut down and start the database, partitioning works > - The experiment was repeated on the second server with 6 databases > running - same result. > > Is it a feature of AIX? > > Thanks > Sev I wouldn't consider any anomaly which would allow you to overwrite a running executable image a 'feature'. Data General provided the same 'wonderful option' of overwriting running executables with recompiles and relinks, but, of course, the running application would segmentation fault and dump core as it tried to read the now-reset memory pointers. I am surprised Oracle didn't do the same. Modifying any executable while it is running is foolish, in my opinion, and anyone who knowingly does so is simply asking for trouble. Granted, the first time you didn't know Oracle was still running, so it was a mistake, and could have been a quite costly one; the second time you were fully aware of what you were doing and, yet, you did it anyway. That you suffered no *apparent* damage was sheer luck and nothing more. You still have no true idea if you DID create internal problems with this act; all you *do* know is problems which could have been caused by this unsupported patching methodology have yet to surface. I wouldn't plan on modifying the accepted and supported patching method any time soon as I expect Oracle Support will take a dim view of this type of exercise. You could easily lose support because of your actions. David Fitzjarrell

fitzjarrell@cox.net wrote: > I wouldn't plan on modifying the accepted and supported patching method > any time soon as I expect Oracle Support will take a dim view of this > type of exercise. You could easily lose support because of your > actions. Easily lose support? I don't think so as long as the customer is willing to pay it. Getting advice from oracle support to recover the database from before the patching attempt along with a re-install. That sounds more likely.

Vsevolod Afanassiev wrote: > Somehow it appears to be possible to patch Oracle on AIX 5.3 without > shutting down the > databases - and everything works fine! > - We have several AIX 5.3 servers running Oracle 9.2.0.7.0 Enterprise > Edition > - Partitioning Option wasn't installed > - I installed partitioning option by running 9.2.0.1.0 Installer, then > 10g Installer then comes with 9.2.0.7.0 Patch Set, and then Patch > 5496862 using OPatch - and then realised that the database was up. > - Everything seems OK, there were no errors during installation, > timestamp on the file $ORACLE_HOME/bin/oracle shows that it was > re-created, I can shut down and start the database, partitioning works > - The experiment was repeated on the second server with 6 databases > running - same result. > > Is it a feature of AIX? In general, when you run an oracle executable under unix, it starts up and then stays in memory. If the executable goes away on disk, the original continues to run in memory. If open files go away, oracle still accesses them because it is using program file handles from when the program started rather than descriptors from the filesystem - so they still don't go away for real until oracle closes them. The problems occur when something gets read in from disk, and doesn't agree with what is in memory. Unless you take the time to figure out everything that is running and called into memory for all your applications, you can't predict what will happen. Remember, when you patch, there are several different things that can happen - you can replace scripts, you can replace or rebuild executables, you can update tables, you can update procedures, you can change file protections, etc. It is impossible to properly test all of these things with a different procedure than the documented one (in fact, stuff gets missed in the much larger regression testing that Oracle does). So your "everything works fine!" statement is suspect. You may indeed be lucky. Do you feel lucky? If you were on some other platform, you probably would have had some part of the install or patching mess up with file locking issues. See catpart.sql, anyways. Part of your luck is probably due to not having already used partitioning. jg -- @home.com is bogus. http://www.henryjenkins.org/

hpuxrac wrote: > fitzjarrell@cox.net wrote: > >>I wouldn't plan on modifying the accepted and supported patching method >>any time soon as I expect Oracle Support will take a dim view of this >>type of exercise. You could easily lose support because of your >>actions. > > > Easily lose support? I don't think so as long as the customer is > willing to pay it. There are instances where a vendor has refused support for customers who do foolish things and expect the vendor to bail them out. > > Getting advice from oracle support to recover the database from before > the patching attempt along with a re-install. That sounds more likely. > -- Michael Austin. DBA Consultant