Veritas-bu

[Veritas-bu] OS Restores in Veritas

2002-06-28 13:18:21
Subject: [Veritas-bu] OS Restores in Veritas
From: dfdwyer AT tecoenergy DOT com (Dennis Dwyer)
Date: Fri, 28 Jun 2002 13:18:21 -0400
Chris,

Depending on how you have your UNIX servers set up I might be able to point you 
in some direction (I didn't guarantee it was the right one). For our NT server 
restores we ran into a similar situation. And yes I believe Bare Metal Restore  
will solve a good portion of these issues once you get over the shock of the 
cost and acquire it. But look at the underlying notion of Bare Metal Restore 
(for NT anyway). Create a system from scratch that has a temporary disk (E:) 
along with C: and D: (that's what we have). Build the basic O/S on the E: drive 
and boot from it. Load the NetBackup client and use NetBackup to restore the 
original C: and D: drive. Reboot off the C: drive and delete the E: drive and 
recover that space back into D: Very simplistic and in a nutshell for sure but 
we've been doing this long before we ever heard of Bare Metal Restore. I built 
my NBU master servers on Sun E450s. I also have Veritas Volume Manager so I can 
set up a mirror of my O/S and root drive and boot off them if necessary and 
hopefully accomplish the same thing that we did in the NT world. Sounds 
plausible but I haven't tested this yet.

Just my two cents worth.

Regards,
Dennis

Quote: "Time is not a test of the truth"
Translation: Just because you've always done it that way, doesn't make it right

Dennis F. Dwyer
Enterprise Storage Manager
Tampa Electric Company

(813) 225-5181  - Voice
(813) 275-3599  - FAX

Visit our corporate website at www.tecoenergy.com

>>> "Christopher Jay Manders

" <CJManders AT LBL DOT GOV> 06/28/02 12:08PM >>>
Hi,

Ok. So I have some questions and have had problems relating to the restore of 
OSes. I have heard alot of input and wanted to summarize so that I can get 
further suggestions and also make some comments.

We have just had our first restore need for an entire OS, and have found a 
number of methods either are not workable or are just plain a pain. Or, perhaps 
we are not approaching this correctly. Commentary welcome. ;)

While it would be nice to purchase Bare Metal Recovery, it costs $1000 per 
system and is not sold except with a $10k consultancy fee. While that might be 
complete boot-able recovery, just installing the OS and running some bootblocks 
writing utility (installboot/lilo) should suffice for the majority of desktop 
users. Perhaps the backup software (Netbackup Datacenter) will do this already. 
The Bare Metal does sound like something worth doing for servers that need to 
be back up in minutes, if possible. But the idea of spending $1k per system out 
in the lab community here (~5000 poeple), when each person gets their own 
funding and funding is tight, is simply a non-starter. Yet the community is 
definately asking about how to do this. 

Anyway, $1k is way too steep a price for us to ask our customers to pay, since 
many are really there only temporarily. I think the idea is that while alot of 
folks will purchase this solution for their servers and true DR needs, desktop 
users just won't pay that much. 

For example, there is a UNIX group here, which we are affiliated with 
internally. They have customers that are looking for some type of solution. One 
just had a very frustrating time with Veritas Netbackup (he was the first to 
even try that here) to try to restore an OS to a Linux box. I actually takes 
some interesting comments from his emails and have highlighted them below.


So, as far as I can see there are the following alternatives for OS installs:

1) Backup up the OS disk in raw partition mode by adding the raw device name to 
the includes_list. The question becomes, is the 'unmounted' part necessary? (I 
mean, with ufsdump, and just plain dump in UNIX, it is highly recommended that 
you do not alter your system or have anything runnign when backups are being 
done. Yet, ufsdump does work quite well for a restore of the OS even then, but 
does truncate some files.) Simply put, if backups in raw partition mode happen 
when folk are not using their systems will it just truncate some files (like 
/var/adm/messages)? Can the raw partition backup function like this?

2) Do a ufsdump/dump of a system disk to a file on a large system and then back 
up that file.

3) Restore the OS from a CD or other media, install the bp client stuff, and 
then restore on top of this everything captured. Now the question is, what 
about ld.so? When that file is overwritten the system _will_ crash, since 
almost everything uses ld.so. And, since there is no networking by default in 
single user mode, this method seems fraught with potential problems.

Here is a UNIX gruy trying to recover that Linux box:
Friday late afternoon:
I'm quite perplexed by the errors from the restore tool.  You can have a
look at bplog.rest.* in /root.

.001 is from the 5/31 incremental -- ignore
.002 is from the true image restore -- most things "were not in the true
image" and were ignored.

Starting at .003, you'll see a bunch of "xyz was not restored";
basically somewhere there was a "Write interrupted by SIGPIPE" and
that's all she wrote.

I ran it again, same problem (.004).

The errors were with /usr (some parts of /usr made it, some didn't).  So
I ran only /usr restore (.005).  Still about 20000 files failed.

Finally, I broke down the subdirectories that failed (about 9 of them)
and ran those again (.006).  I didn't get any SIGPIPE or "was not
restored" messages, but the final message says:

18:04:39 (71047.xxx) INF - Status = the restore failed to recover the
requested files.

So I'm not sure what happened here.

 

4) Restore to another area and copy the restored files back to the original 
system. So, potentially have a restore machine with a huge amount of disk space 
that you restore everything to. Then you copy that to the disk that is mounted 
somewhere after first being booted with an OS with theVeritas Netbackup client.

Monday:
The restore to tortoise does not work.  There are apparently still some files
missing in usr as X fails to start up (it does really weird things and hangs).

Is it possible to have you restore /usr from the XX restore somewhere else and
see (a) if it has errors restoring like I did (See email sunday) and if it
does not, (b) I can then copy it to tortoise /usr and try that?

This method appeared to get most everything, but here is what he says last:
I'm trying to make the restored system work but have the following problems
with links and devices:

1) absolute symlinks are relative to the path on servback (/opt/logs/Restored)
- this is easy enough to fix with a find/perl script.  However, for future
reference, the answer to the symlink question should be no (when initiating
the restore).

2) all hard links are missing.  I'm not sure if I can recover from this. There
is a question when initiating the restore about hard links that I'm not sure
how to answer.  I left it the default yes.  The hard links are missing from my
restored /sbin as well as the stuff on servback.  I should be able to recover
hard links that are made by rpms by simply reinstalling the rpm.  However, for
apps not installed with rpm that may use hard links, we could be hosed...
3) devices seem to have the wrong group membership. don't know why that is the
case.
I'll keep you posted.  This is something, however, that we need to figure out;
especially #2.

 
So, is this the method that are recommeded for an OS install after all? And, is 
using the 'bp' and 'bpadm' utilities enough to assure that the symlinks are 
correct? He was using the 'bp' utility on the UNIX client, so I am just not 
sure. One is not correct.
> >did anyone successfully restore OS?  I had hard time restore hard links.  I
> >want to restore to an alternative client.  but, no matter whether 'rename
> >hard links'is checked or not, hard links were not restored properly.  So I
> >was not able to boot from restored OS.  Did I miss something?
> 
> Although it's a bit late, here goes.
> 
> The "not restauration" of hard linked files is actually caused by a bug in 
> the java
> gui. It is not passing the ticked option about hardlinks to bprestore. 
> 
> There is a patch for this from veritas. Don't know the number of the patch, 
> but search 
> for java, gui and hardlink. Workaround is to use Motif GUI or commandline. 

So, is this the recommended way?

Also, why the changed ownerships and permissions this guy had?




Thoughts?




Thanks!


Chris






<Prev in Thread] Current Thread [Next in Thread>