We are setting up a new Disaster Recovery environment.
On Production we have SQL Server cluster with two nodes.
For DR we decided that one node should be enough. The plan is to add the second node ASAP if disaster happens.
We requested our managed cloud company to create a single node cluster for us. However, they try to discourage us from doing that and suggest installing non-clustered instance instead, so when disaster happens we will build brand a brand new cluster with two brand new nodes and then migrate to it from non-clustered instance.
This option sounds like more work for us if disaster happens.
We will need to migrate all the databases to the clustered instance from non-clustered instance. I also expect a longer downtime if we do that, as the opposite to just adding a second node to the existing cluster.
Tech person from this cloud company said that single node cluster is complicated solution and is prone to all kinds of problems like “split brain”, that it can stuck all of a sudden and they have seen so many issues with single cluster nodes. He said that Microsoft does not recommend having single node clusters. He wasn’t very specific and I’m not sure I understood what exactly problems we will have.
I tried to google for some specific problems, but failed to find were Microsoft would suggest to avoid having single node clusters, especially when it comes to DR.
Is there a reason to avoid single node cluster solution for DR and go with non-clustered instance instead?
Overall, there is no need for a cluster IMHO. If you need HA at the DR site, then do it – but if you don’t, I don’t see a purpose for it as the stand alone will work just fine. The reason I say this, is that they will be two distinct clusters which makes it a little hard to justify.
Tech person from this cloud company said that single node cluster is
complicated solution and is prone to all kinds of problems like “split
brain”, that it can stuck all of a sudden and they have seen so many
issues with single cluster nodes.
So I don’t know if the tech person didn’t understand, or what, but you can’t split brain with a single node… so, I don’t know why (s)he thought that, but that’s incorrect. There really aren’t any issues with a single node cluster other than it’s a single node cluster which defeats the point of having a cluster TBH. It’s a different story if it’ll eventually be built out, but if you’re going to wait for a disaster to do so… you’re already behind the ball.
Keeping just a single instance, not in a cluster would be faster to replicate as well should something happen.
In the end it’s up to you, unless I had a much more compelling reason I’d go with just a single server without clustering. Also, don’t listen to their techs anymore 😉