[cisco-voip] Bridge upgrade vs. sub(s)

[Beck, Andre]

Hi,

I'm preparing for an upgrade of a cluster (just one sub) running 7.1(3)
on DL380G4s (7845H2) to 8.6(2) on DL380G6s. To make things interesting,
the old version doesn't even recognize the new servers, and the new version
is not supported on the old metal except for bridge upgrade. Needless to
say, the upgrade has to be done with minimal downtime, something like
15min to switch would be acceptable, not more.

I'm planning to proceed as such:

1) Shut down the pub, while the sub continues serving the live network.
2) Boot the pub using GRML or such Linux live CD and make a full disk
   copy to external storage. Maybe also break the mirrors and insert
   additional disks to have a physical copy through RAID, putting the
   original disks at a safe place.
3) Isolate the pub in a replica of the original network including
   default gateway, DNS and NTP servers as needed. Boot it up again.
4) Do the upgrade procedure to 8.6(2) as documented, ending up with
   a bridged upgrade only machine.
   *** Here the docs state I have to also do this to the sub(s). Of
   course I can't, they are serving the live network here. Is this
   fatal, or is my procedure a proper workaround?
5) Do a DRS backup of the bridge-upgraded pub.
6) Restore the original pub using the copies made in step (2), plug it
   back into the live network, boot it up and have the live network
   working fully redundant again.
7) Install 8.6(2) on a new DL380G6 plugged into the replica network.
8) Restore the DRS backup from (5) to this new pub.
9) Make sure the pub is working correctly, except for the missing sub.
10) Now add a second DL380G6 to the replica network, install 8.6(2) as
    a sub on it, and reestablish the cluster. Manually make sure all
    files in the TFTP server of the original sub are available on the
    new sub as well, as that one isn't restored from backup.
11) After making a lot of tests and sanity checks, swap in the new
    cluster in a short downtime window.

Does this plan make any sense? Is there a better one? I'm specifically
bothered by the documented need to perform the bridged upgrade on all
nodes in unison, and then to restore first to the pub, then to all the
subs. Of course that's impossible without inacceptable downtimes. But
I assume that I cannot, after step (6), just isolate the old sub in
the same way as the pub before, doing a bridge upgrade of that isolated
sub only to get hold of a DRS backup of the sub?

Any hints or ideas welcome,
TIA,
Andre.
-- 
                    Cool .signatures are so 90s...

-> Andre Beck    +++ ABP-RIPE +++      IBH IT-Service GmbH, Dresden <-