April 13, 2005 at 9:46 am
Just was wondering if someone could verify if the following was possible and verify what extra/special steps are required on top of a standard active/passive SQL cluster install.
We want to have two databases, one production and one test database. Both of these databases will be clustered but they will share the ONE passive cluster.
e.g. if production db AND the test db failed they would both failover to the same shared passive db server.
The chances of both databases failing over are EXTREMLEY rare but am happy to accept this if it ever occurs.
I am guessing that firstly you would setup a normal active passive cluster (e.g. for the test platform) but I am trying to understand how the second db server (production) is configured and installed. Will there be TWO seperate sets of SQL server binaries on the passive server?
Any help would be greatly appreciated.
Thanks,
TD
April 13, 2005 at 2:07 pm
How many physical servers are you talking about here, 3? Or are you talking about just 2 physical servers with both instances running on 1 server while the other sits idle?
Any case, you can only have one cluster here. You can say run this instance of SQL Server on Node1 and Node3 and run that instance on Node2 and Node3, which sounds like what you're doing. If that's the case then there will be two sets of binaries on that Node3 server. Also, only one of the instances can be a default instance. The other wil have to be a named instance.
K. Brian Kelley
@kbriankelley
April 14, 2005 at 8:22 am
There would be three physical servers
Server 1 - Test DB
Server 2 - Prod DB
Server 3 - Failover server for both Server 1 and Server 2
As you mentioned and as I guess we would have two separate sets of SQL Server binaries on Server 3 and both the databases would need to run on different ports. How do I go about setting this up?
Do I setup Server 1 and 3 as a normal active passive cluster. Then create the additional cluster resources required by the second database instance and add Server 2 as a third node. With Server 2 then set as the primary do I run a second install of SQL server in which a second set of SQL server binaries will be installed on Server 3.
Does this make sense?
Thanks,
TD
April 14, 2005 at 8:45 am
Hi,
you might even use a active/active configuration, with 2 instances of the sql server on different ip adresses, 2 cluster groups, 2 Disk groups. normally each instance would run on its own server, but could be switched to the other one.
a 3 node cluster would essentially be the same, you just would have to configure different preferred nodes...
(that's just a guess, have not done something like that yes...)
karl
Best regards
karl
April 14, 2005 at 9:23 am
You should go ahead and set up on the OS side a 3-node cluster. When you go to install SQL Server 2000 in the cluster environment, it'll prompt for what nodes it should be running on. The first instance (production), you could install as the default instance, meaning it'll listen on tcp/1433. When you go to install the second instance (meaning a second install), it'll have to be a named instance. Just as with a named instance on a single server, it'll automatically generate a random port to use but you could set it to what you want (just not tcp/1433 for obvious reasons).
K. Brian Kelley
@kbriankelley
April 21, 2005 at 1:20 pm
Now wait a minute. First off, are we talking Windows 2000 or 2003? That makes a big difference. For Windows 2000, unless you're running Data Center Edition, aren't you limited to 2 nodes in a cluster? If that's the case then you'd need a named instance like Brian said. And even then, you don't have a passive node unless both instances are on the active node (not a pretty picture believe me - I lived through that).
For Windows 2003, your instances can all be default with different SQL Server names and VIPs, can't they? That's what I've been planning on for server consolidation (I have four clusters that I'd like to collapse into a 4+1 cluster, freeing up 3 8-way servers).
I haven't gotten onto 2003 clusters yet - funny how hardware is scarce for testing/proof-in-concept, but how quickly it shows up when they've already sold the product and need it implemented.
April 22, 2005 at 5:16 am
I think the node limit was removed in a sp for w2k.
The main limit for w2k is memory your active nodes would be limited to 4GB ( your inactive node would need node1 mem+node 2 mem ), with w2003 you can up this to 32Gb (total ) giving a much better implementation.
[font="Comic Sans MS"]The GrumpyOldDBA[/font]
www.grumpyolddba.co.uk
http://sqlblogcasts.com/blogs/grumpyolddba/
April 23, 2005 at 2:05 pm
I am talking about a Win 2003 server. At the moment I am planning to try it with two separate instances on SQL Server with the passive server having two sets of the SQL server binaries.
April 25, 2005 at 1:06 pm
Brian Kelly's response is dead-on. I just created a 3-node cluster just as you are proposing, and it was a snap. I would suggest going to the Microsoft website and printing out the failover cluster installation instructions, including the pre and post installation worksheets. If you follow these, you will not have any problems implementing the configuration that you are describing.
Viewing 9 posts - 1 through 8 (of 8 total)
You must be logged in to reply to this topic. Login to reply