[1599] in SIPB-AFS-requests
Re: Cutover plans
daemon@ATHENA.MIT.EDU (Derek Atkins)
Tue Nov 15 18:37:44 1994
To: Greg Hudson <ghudson@MIT.EDU>
Cc: sipb-afsreq@MIT.EDU, yandros@MIT.EDU
In-Reply-To: Your message of "Tue, 15 Nov 1994 18:13:37 EST."
<9411152313.AA00721@glacier.mit.edu>
Date: Tue, 15 Nov 1994 18:34:48 EST
From: Derek Atkins <warlord@MIT.EDU>
Wait -- we're moving both rosebud _AND_ ronald-ann at the same time?
And to move ronald-ann we're just going to unplug it and then replug
it in? Ouch -- that means we're going to have to rebuild the vldb!
> What's the best way to deal with this if we don't move the RO volumes
> during the swap? vos remsite the RO volumes before the move and vos
> addsite/vos release them after? I guess that will work, without
> creating an outage.
Yea, that's exactly what I was thinking of doing. However if we have
to rebuild the vldb, it really doesn't matter, since rebuilding the
vldb will take some downtime anyways.
So, here is what I see us doing (and it doesnt require any temp-space
at all!). This will not only be the quickest way of doing it, it will
require (imho) the least amount of downtime as well. What do you
think of this plan?
1) put the new machine up as new-rosebud, setup as a fileserver only.
2) start moving all the RW data off rosebud and onto new-rosebud
directly.
3) re-add the RO volumes to new-rosebud
4) Once all the volumes have been moved off rosebud onto new-rosebud,
make sure that no volumes remain on rosebud.
Once there are no more volumes on rosebud, then we have our short
outage:
5) Bring down ronald-ann, and move it to 18.181
6) Bring up ronald-ann as a db server, talking to the old
Rosebud (which should know about the new ronald-ann at this point)
7) destroy the vldb database on old-rosebud and ronald-ann, and then
rebuild the vldb (syncvldb) from ronald-ann and new rosebud.
At this point, people will be timing out contacting old ronald-ann,
but will still win with old-rosebud, and the servers should be ok.
8) add new-rosebud as a db server to the other machines and then
create the db processes on new rosebud
-derek
