Re: [Networker] Need help with migration test
2009-12-02 15:17:07
Francis Swasey wrote:
I moved from 64-bit Solaris (Solaris 9) to 64-bit Linux (RHEL 5) in May
of this year. I did not change versions (I was running 7.4.3 on both
servers). I did not change host name or address.
My general process was to follow the DR guide (mmrecov) and then use
nsradmin to make the changes to the (mostly) notifications paths for the
OS differences. We contracted with Cambridge Computer Services
(http://www.cambridgecomputer.com) for a procedure and a few hours of
assistance.
It is most certainly possible to make the change from Solaris to Linux
without having to scan (in my case) seven years worth of tapes --
despite what your EMC sales/support folks will tell you.
The /nsr/indexes directory is endian neutral and therefore is just an
rsync from old server to new server to move it.
Thanks,
Well, I spoke with tech support, and one crucial point I forgot to
include is that in our case because the host names between our old
server and new server are different, even if we rsync the indexes, we
will still encounter a problem later down the road if and when we ever
need to restore or merge in an older index, using 'nsrck -L7', that was
backed up on the old server. At that point, it will recognize that these
index(es) do not belong to the current server, and we will be denied.
This is because the client id on the new server will not match the
client id that the index was saved to when it lived on the old server.
As a result, we'll need to change the client id on the new server to
match the value on the old server. So after running 'mmrecov', we'll
need to change the server's client id. This will then change all the
bootstraps and indexes to now be listed for the new server. We can then
restore (nsrck -L7) the indexes so they will now be owned by the new server.
Of course, you can't change the client id. You have to set its value as
soon as you create the resource, so we'll have to follow the steps for
changing the server name via the "temp name and then back again"
trickery game. I think others have run into this problem before when
trying to use 'nsrck -L7' to recover or merge in indexes that resided on
another server, never mind the fact that the new media database does
list the required volume(s).
Obviously, we could run scanner or saveset recover, but if we want to
merge in older entries, we'll need to use 'nsrck -L7'.
George
Frank
On 11/30/09 9:48 PM, George Sinclair wrote:
Hi, moving an older 64-bit NW 7.2.2 Solaris server to Linux server
with different host name. For now, we're running in test mode on this
new server, so wanted to check the migration steps as I've listed
below. Also, I have several questions/concerns further down.
I've read through the disaster recovery guide (7.51) and looked though
old postings here, but I'm still confused, so I'd greatly appreciate
any help. We have way to many tapes to scan everything in, even
rebuilding indexes would be a gargantuan task.
Specifically, we've done a few full backup tests on the new server for
some existing/older clients, wherein I created the NSR client
resources from scratch, but we'd like to perform some recovery testing
via saveset recover and nwrecover from some volumes written on the old
server. I'd like to just play around with a few clients for now and
then do a final migration later once I know this all appears to be
working.
BACKGROUND: First, we're not ready to go live yet, so I know we'll
need to redo everything once we're done testing and dedicate one day
to getting the media database and indexes re-migrated by COB that day,
Also, we would really prefer not to upgrade the old server. We'd like
to just leave it alone since an upgrade is one-way and we have little
space on there, and we'd like to be able to go back if we had to.
Alternatively, installing the same, older version on the new server
first, and then subsequently upgrading it, is not really an option
either. But the plan would be that after testing, we get the host
transfer from EMC, do the final migration and then shut down the old
server once we're ready to go live.
Also, I don't really care about any of the non-Legato nsr related data
on the old server. So, I anticipate making the old server a client of
the new server for maybe just one final full and that's it. I don't
know that we would even need the old server's index on the new server?
OLD HARDWARE: Old server (Solaris 9) uses two RH Linux snodes, each
with its own tape library (one STK LTO-3, the other a Quantum
SDLT600). No tape library directly attached to old server, so we use
file type device to back up server's bootstrap and index, and those
are automatically cloned to LTO-3 and staged (different clone pool)
every two weeks to LTO-3. The media database is very small (320 MB),
and total indexes (approx. 34 clients) are less than 25 GB.
NEW TEST HARDWARE: RHEL box (64-bit, Dell PowerEdge 1950) running
32-bit NW 7.5SP1 with latest version of NMC. Way faster box, way more
disk space. This test server has a *different* host name and uses a
Dell PowerEdge R905 Linux snode (32-bit NW 7.5SSP1) with attached Dell
LTO-4 tape library.
Since the old/new servers have different host names, we can have both
up and running NW at the same time. We would prefer not to change the
host name on the new server, however. Thus far, we've configured the
LTO-4 library and backed up a few existing 7.x clients that are
currently being backed up by the old server by changing
/nsr/res/servers on those clients to list both servers and
stopping/restarting client software. Everything looks good so far.
<<< MIGRATION STEPS >>>>
On old server:
1. Obtain client id for server
mminfo -q 'savetime>yesterday,client=server' -r clientid | sort -u
2. Clone latest bootstrap to a new clone pool volume.
note: this will be the only save set on this clone volume.
note: this is kind of cheating because I know I should rerun a level
full of the bootstrap as: `savegrp -O -c server`, since the media
database will have changed since the last bootstrap, but I just wanted
to save time for this test. For the final migration, natch.
On new server:
3. Set up symbolic links locations for /nsr so the directory structure
is the same as the old server:
Old server: /nsr -> /disk2/nsr
/disk2 -> /0/disk2
note: /0 is the file system, and disk2 is a directory under /0.
New server: /nsr -> /opt/nsr
a. Shut down networker software on new server
b. `rm /nsr`
c. `mkdir /0; mkdir /0/disk2`
d. `cd /0/disk2; ln -s /opt/nsr nsr`
I'm not sure about this step, but I'd read somewhere that mmrecov will
not work if you don't preserve the nsr directory structure?
4. Restart NetWorker.
5. Remove NSR client entry for server. Recreate NSR client entry,
giving it same client id as old server.
6. Manually load clone volume from step 1 into device on new LTO-4
library and run:
mmrecov /dev/nst#
note: since this will be the only data on the tape there should be no
reason to enter starting rec/file numbers, just hit return?
7. Shut down NetWorker on new server. Rename /nsr/res to /nsr/res.org.
Rename /nsr/res.R to /nsr/res.
note: I guess /nsr/index will remain as is and /nsr/mm will be
overwritten? I don't really care about /nsr/mm because we've only run
a few test backups on the new server (just one tape), so I can always
scan that in later.
8. Restart NetWorker.
9. Recover a few client indexes as:
nsrck -L7 clientname -t date
Not sure if this will work due to the fact that these were from a
different server, but maybe step 5 will permit this? If so, it should
request volume containing last full and then latest level 9. Could
just have created level fulls of the indexes to begin with ('savegrp
-O'), but thought instead that this might be a good test to see if the
new server can use the media database? We're only recovering a few
indexes to test, anyway.
10. nsrck -X
QUESTIONS/CONCERNS:
1. It's my understanding that the media database is not endian neutral
so disaster recovery must be used, but the indexes are neutral so they
could be copied/rsynced. But I've read that it's recommended to still
recover from tape. That true? Guess it's a good test regardless. I
know EMC does not officially support OS platform migration, but others
have done this successfully ...
2. Most of our clients could conceivably start from scratch on the new
server, and we could always run 'nsrck -L7' to merge in older entries
later if need be, BUT we don't have a large number of clients, and the
total size of the indexes is rather small. Also, we do have a couple
that we absolutely will require maintaining the index and picking up
where we left off as soon as we go live - at least until we can run
new fulls on all their save sets - and those few clients have a large
number of very huge save sets with a long browse policy (most of the
save sets do not change very often), so it would take us two weeks to
run fulls on all that. As a result, we'd like to keep those indexes so
we can run fulls at our leisure.
3. Can we in fact use the recovered res directory? Will the mmrecov
re-create all our NSR client resources, group, schedules, etc? OR will
it be necessary to re-create all that manually and instead use only
the recovered media database?
If so, I *can* do that as I wouldn't have to recreate all that much
for some simple initial tests. I could finish re-creating everything
in probably a week, if necessary, and periodically repeat the above
steps to get the latest media database info. Obviously, if I have to
do this manually, though, then I guess 'nsrck -L7' won't work until
I've recreated at least one NSR client resource for each affected client.
4. Will the recovery of the res directory complain about
jukebox/device configurations that are not known to the new server?
Once we're live, we're planning to move the SDLT tape library to the
new snode, and reconfigure and then install NW 7.5SP1 on the other
snode running the LTO-3 library, but we can't do that until then since
we still need those for our production backups on the old server.
We're not so concerned about the LTO-3 library, however, since we can
read those tapes on the LTO-4 library.
5. What about the server license? Even though we have a different host
name on the new server, will it disable itself after we recover the
old media database? I can add a temporary enabler code? we cant do the
host transfer until after we've had a change to test more. I mean,
once we do that, we need to be ready to roll, and we're not there yet.
6. Should I expect an MD5 checksum listing of an index directory on
the old server to be the same as the recovered version on the new
server? Or, will NW modify the whole thing, making it moot to even
compare them?
7. I was thinking to run both a media query and a save set query of
the whole database on the old server before the migration and compare
this with the same queries on the new server, following the recovery,
but not sure if spacing or other format issues might produce different
results?
Also, now that I think about it, if I don't run a fresh full of the
media database (step 2) then I guess I can expect there to be
differences with the differences being whatever was backed up since
the last bootstrap?
Thanks!
George
--
George Sinclair
Voice: (301) 713-3284 x210
- The preceding message is personal and does not reflect any official or
unofficial position of the United States Department of Commerce -
- Any opinions expressed in this message are NOT those of the US Govt. -
To sign off this list, send email to listserv AT listserv.temple DOT edu and type
"signoff networker" in the body of the email. Please write to networker-request
AT listserv.temple DOT edu if you have any problems with this list. You can access the
archives at http://listserv.temple.edu/archives/networker.html or
via RSS at http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
|
|
|