We just went through a kind fo painful upgrade from a Windows 2000/SQL 2000
SP3 2 node cluster to a Windows2003/SQL 2003 SP3 2 node(new servers) cluster
upgrade. Everything I had read pointed to this being very simple, stating
evit one node, add the new server, install SQL, install SP3, failover and
start over on the other node. We ran into some problems I would like cleared
up.

1. Proper procedure for installing SQL SP3a?
I tried installing it on the failover node and not the server that was
active. This was aparently a mistake because you can't install SP3 on the
inactive node. When adding a new node, do you have to a apply the SP3 to the
active node again, even if it is already installed to get the SP applied to
the inactice nodes?

2. DTC.. This was a pain. When first installing and following all the proper
KB articles, it would not start. Finally in the MS clustering book we found
had it mentioned creating MTC virtual IP and virtual name reasources, adding
dependancies then it would start. Has anyone else done this in a
W2003/SQL2000 cluster?

Thx!

Comments Inline

--
Geoff N. Hiten
Microsoft SQL Server MVP
Senior Database Administrator
Careerbuilder.com

I support the Professional Association for SQL Server
www.sqlpass.org

"msnews.microsoft.com" <jealy (AT) omniumworldwide (DOT) com> wrote


There is no SQL 2003 product. I assume you meant SQL 2000 SP3a

You must first install SP3 (or SP3a) from the node hosting the resources.
The install will update binaries on all reachable nodes during the upgrade.
You can apply a service pack to the non-host node ONLY when you are
replacing a failed node and you need to update just the local binaries.

Since you had Windows 2003 available, I would have upgraded the existing
hardware in-place (one node at a time) You can then add the new nodes and
remove the old ones. Windows 2003 and SQL 2000 support up to 4 nodes in a
cluster. (Yes, Windows 2003 will support 8 nodes, but SQL will only install
to 4.) You would have then been able to do the entire upgrade with minimal
downtime. I have done this in the past and it worked succesfully.