[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference eps::oracle

Title:Oracle
Notice:For product status see topics: UNIX 1008, OpenVMS 1009, NT 1010
Moderator:EPS::VANDENHEUVEL
Created:Fri Aug 10 1990
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:1574
Total number of notes:4428

1547.0. "Upgrading OS - what does OPS do??" by DYOSW5::WILDER (Does virtual reality get swapped?) Thu Apr 17 1997 12:06

    Here is an interesting situation:
    
    We need to do a rolling upgrade of DUNIX in a cluster that is running
    OPS. The documentation states that the node on which the upgrade will
    happen must have it's instance shut down, thereby leaving Oracle up on
    the remaining node.
    
    Now, in order to do the upgrade of the OS, I must delete the cluster
    member. When I do this, the surviving member must modify the drd
    services (we are using preferred service with all the node names listed
    in our preference). When ASE modifies a service, it deletes the service
    and then adds it back in. When it adds the service back in, it sets all
    the ownerships on the drds to root.system.
    
    My question: If Oracle OPS is running on a single surviving node, what
    will happen if the drd services disappear, and then re-appear with the
    wrong ownerships (not oracle.dba as is required)? Will Oracle have some
    heartburn? Is this a workable situation? Or, must I shut down Oracle on
    the surviving member while the cluster re-configures itself and we
    reset the ownerships by hand? If this is true, it negates our claim of
    "never having the database go down while upgrading individual nodes".
    
    Any ideas on this?
    
    Thanks,
    
    /jim
    
T.RTitleUserPersonal
Name
DateLines
1547.1UTRUST::PILMEYERQuestions raise the doubtFri Apr 18 1997 04:5823
    One way around this is to do what I did. But it has some adverse
    effects also.
    
    Instead of using the /dev/rdrd/... device specials I had oracle using a
    NFS shared directory in which I created the drd device specials (using
    the supplied script) and make symbolic links to those. The symbolic links
    would have the names of the underlaying LSM volumes and suddenly
    everything got a lot more understandable. I wrote a script that did
    this automatically including setting the device specials to the correct
    protection.
    
    When the service is modified the cluster software never touched my
    device specials. Which helps in the case you're investigating. Of
    course when you change the drd configuration you have to remember to
    run the script.
    
    The bad news is that on an upgrade the major number for the drd's can
    change (it did on the 1.0 to 1.4 upgrade) and then you're using the
    wrong major number on the systems that have been upgraded (or the other
    ones). In fact the major numbers can even change on a reboot. But there
    are ways around that too.
    
    -Han