I'm putting the finishing touches this week on a perl script
that handles Oracle database hot backups (to keep the database
up and running 24/7). It's very much like the tcl script John
Jackson wrote, but uses the Expect perl module to interact with
the database.
The script essentially backs up all of the tablespaces, control
files, and config files to a spare partition set up specifically
for this purpose. This partition is then set to be backed up
with Amanda, along with all the regular partitions on the
system (e.g., /usr, /var, etc.).
So if you'd like to look at a perl version of this, let me know
and I can ship a copy your way.
-Original Message-
From: Fredrik Persson P (QRA)
[mailto:[EMAIL PROTECTED]]
Sent: Monday, November 20, 2000 8:46 AM
To: '[EMAIL PROTECTED]'
Subject: RE: Backing up Oracle database
[Fredrik Persson P (QRA)] Please read the entire mail!
In the end, I'm coming to a conclusion that
probably obsoletes a lot of the previous junk that I
wrote. I've been writing this email all day and
I've become wiser along the day too...
I've been assigned the task to back up an Oracle database
w/ amanda. I'm
also told (by the db administrator) that a have to read
the files under /u02
(a directory in the root of the file hierarchy) in a
certain order.
Ummm, that's very odd. I've talked with my Oracle admin
and we're not
sure what it is you're trying to do here. Could you elaborate?
[Fredrik Persson P (QRA)] Absolutely! Our main concern
seems to be that the
database must operate 24/7. The scheme that we thought
up was to run the "begin
backup" command on all tablespaces before amanda starts
her work. When she's
finished, the "end backup" command is run on all
tablespaces. When all the tablespaces
are brought "back into action", the command ALTER
SYSTEM SWITCH LOGFILE is issued,
to create a new ARCHIVELOG file, even thought the old
one might not be "full".
To restore the contents of the db, we first bring the
tablespaces back, and then apply
the archivelog file to them. Now, there seems to be a
file called the "control file". If this
file is backed up before the tablespaces, and the
tablespaces are altered during backup,
we cannot restore the system. The control file is out
of date and we get an error message
similar to "control file older than tablespace". Can't
recall the exact words, but that's what
it means.
So, what I want to do is to backup the tablespaces
first, and the control file after that. That's
why I need to tell amanda to do things in a certain order.
Also, the way we back up our Oracle system is to dump it to
a scratch disk
area and then let Amanda back that up. And yes, we've done
restores :-).
And they even worked :-).
[Fredrik Persson P (QRA)] Are you running 24/7? If so,
how do you deal with the issue
of changes in the db during file transfer? Do you use
raw partitions or does Oracle run on top
of a file system?
The scripts we use are at:
ftp://gandalf.cc.purdue.edu/dbbackup.*
[Fredrik Persson P (QRA)] That link didn't work. Did
you mean ftp://gandalf.cc.purdue.edu/pub/amanda/dbbackup.*
Is there a way to tell Amanda this?
If this really ends up being the way you need to do this, I
have some
hacks that I think would work. But let's make sure you're
headed down
the right path first.
[Fredrik Persson P (QRA)] After spending the day
pondering this issue, I think we've come to the conclusion that
we'll copy the db files to a temp storage, and then
backup them from there as if they were regular files.
This is your approach too. I'd say it's a good one.
Minimizes db backup time and gives us control
over the order when storing db files.
I've browsed your scripts. Are those GPL'ed or
something similar? Right now, we're using our own scripts
that can be said to be stripped versions of yours.
However, I might be interested in your stuff later.
Thanks for your help! Using a temp storage was pretty smart.
/Fredrik Persson
