Updating database servers using dbnodeupdate.sh - part 2

• The Exadata Database Machine Owners Guide (Chapter 7) has
  instructions to create backups stored outside of the dbserver, for
  example on an NFS mount (see 'Creating a Snapshot-Based Backup of Oracle Linux Database Server')
• dbserver_backup.sh - which creates a local copy of your active lvm.
  

In this post I will explain background and usage for both backups and how they integrate with dbnodeupdate.sh

dbserver_backup.sh

For backing-up and rolling-back Exadata dbserver OS updates  the
dbserver_backup.sh script is used by dbnodeupdate.sh. For each upgrade
by default the dbserver_backup.sh script is executed. When executed (either manually or via dbnodeupdate), the
dbserver_backup.sh script creates a small snapshot of the 'active' sys
lvm. The active sys lvm is the primary lvm that your current OS image is running on. For example:

[root@mynode ~]# imageinfo


Kernel version: 2.6.39-400.126.1.el5uek #1 SMP Fri Sep 20 10:54:38 PDT 2013 x86_64

Image version: 11.2.3.3.0.131014.1

Image activated: 2014-01-13 13:20:52 -0700

Image status: success

System partition on device: /dev/mapper/VGExaDb-LVDbSys2

In the above example the active lvm is /dev/mapper/VGExaDb-LVDbSys2.The snapshot is created to have a 'consistent' 'view' of the root
filesystem while the backup is made. After the snapshot is created, it's
mounted by the same script and then it's contents are copied over to the
inactive lvm. For lvm enabled systems, there are always 2 'sys' lvm's "VGExaDb-LVDbSys1" and
"VGExaDb-LVDbSys2". VGExaDb-LVDbSys2 will automatically be created (on
lvm enabled system) if not existing yet. For the example above, the 'inactive' lvm will be VGExaDb-LVDbSys1


Now, depending on how many files there are in the root (/) filesystem
(based on your active sys lvm) the backup times may vary. Previous Grid and Database home installation zip files in
/opt/oracle.SupportTools/onecommand will make the backup take longer (not
the restore, which I will explain why later). Same for those who have many small files (like mail messages in /var/spool) - the backup may take longer. 


One of the first steps the dbnodeupdate.sh script will doing when executed is making a backup with this script. Now, if you want to shorten your downtime and make this backup before
the start of your 'planned maintenance window' you have 2 options: Either execute the dbserver_backup.sh script yourself or use
dbnodeupdate.sh with the "-b" flag to make a backup only before hand.


Example making a backup with dbnodeupdate.sh here (see 'Backup only' for 'Action')


When you then have the downtime for planned maintenance and already have
the backup you can then let dbnodeupdate skip the backup using the "-n"
flag.


Example skipping a backup with dbnodeupdate.sh here (See 'Create a backup: No')


Both Sys lvm's are 30GB each. The snapshot that will be created is ~1GB. It is recommended to keep this in mind when claiming the free space
in the volume group to make your /u01 filesystem as big as possible. (the script checks for 2 GB free space in the volume group)


Now, when the update proceeds, the current active lvm will remain the
active lvm. This is different than what happens on the cells where the
active lvm becomes inactive with an update.  Typically you will only
switch active sys lvm's when a rollback needs to be done on a db server, for example,
an upgrade from 11.2.3.3.0 to 12.1.1.1.0 needs to be rolled-back. What
happens then is nothing more than 'switching' the filesystem label of
the sys lvm's, updating grub (the bootloader) and restoring the /boot
directory (backed up earlier also by dbnodeupdate.sh). Then, a next
boot will now have the previous inactive lvm as active lvm.


Rolling back with dbnodepdate.sh as in the example here (a rollback from 11.2.3.2.1 to 11.2.2.4.2) 

After booting the node, it's recommended to run dbnodeupdate.sh again with the "-c" flag to relink the oracle home's again.


Remarks:

• It's important, to make a new backup before attempting a new update.
• In the above example, there is only talk about the sys lvm's. This
  means custom partitions including /u01 are not backed up. For regular
  node updates this is enough to rollback to a previous release but it's
  recommended to also have a backup of other filesystems inline to your
  requirements.
  
• Nodes deployed without lvm will not have this option available
• Rolling back db servers to previous Exadata releases with this procedure does not rollback the firmware