SQL Server clusters are widely used, and widely misunderstood, which makes people like our good friend (and clustering MVP) Allan Hirt cry bitter tears. So here are a few notes for you and yours.
Cluster IS NOT Disaster Recovery
A cluster does not protect against disaster. Unless it’s a very, very, very limited scope disaster (that scope being, the hardware of the one server running SQL, not including the disks). A cluster is considered a High Availability solution.
Why: A typical cluster lives in a single physical area (typically, the server room). And, a cluster shares disks. Anything that goes wrong with the SAN, the power, the building, or the network, goes wrong for the ENTIRE cluster.
Failover IS NOT Ultimate Solution
Failing over the cluster should not be your first reaction to any given problem. Letting your DBA troubleshoot the issue should be the first reaction.
Why: If there’s a problem, you need to know the source of the problem, so that it doesn’t happen again. Not to mention, restarting the service and/or failing over the cluster can exacerbate the problem AND wipe out important troubleshooting data. It’s lose-lose!
Failover IS NOT A Freebie
A failover is (potentially) quicker than a restart of a single server, sure. But you still lose all your connections, all open transactions are rolled back (and not automatically restarted), and you still have to wait for the SQL Server service to spin up.
Why: SQL Server itself doesn’t know the difference between failing over, and simply being restarted on a single instance. It’s the same thing, in SQL’s eyes.
Second Node IS NOT Magical
This one is a two-fer. First, understand that if your secondary node isn’t functional for whatever reason, you don’t actually have a failover cluster. Plenty of shops ignore issues on their secondary node for years, just assuming that they’ll never need it.
Second, moving the services and disks over to the second node rarely gets you anything that a service restart wouldn’t. That second node isn’t magically immune to deadlocks, long transactions, nor memory and disk pressure. Even in cases where SQL Server isn’t responding, or is deeply broken in some way, a service or server restart is at least as good as a failover.
Why Do We Even Have This Cluster?
I don’t know. Maybe someone high up fell in love with the word “cluster”, the way they’re falling in love with “The Cloud” right now. Maybe you have a legitimate need for, say, rolling upgrades with minimal downtime.
Regardless, you have it. Understand the concepts, and understand the few cases in which failing over is actually beneficial. Here, have an Introduction to Clusters, with diagrams. Need more? Check out Allan’s 4-day training classes.
P.S. I’m on a boat this week, and so I’ll be out of touch for a while. The out-of-touchness includes our awesome DBAs@Midnight webshow, which next airs on February 8 at 11pm Central Time. Catch up on back episodes at http://www.MidnightDBA.com
P.P.S More resources, from SQL Server BOL: