May I run DAG replication between two clustered instances in an FCI?

My ambition is to run ASPStateInMemory with durability on, supplementing the in-memory database with an on-disk version to give session state persistence when and if the clustered instance of SQL runs into, say a network card failure and needs to fail to a partner node.

Can I do this?

I understand that DAG's are usually run between standalone instances of SQL running on different machines.

But, can I combine the DAG with the FCI?  Instead of using a remote standalone SQL server, can I just another instance in the same cluster?

Thanks,

Bob.

August 27th, 2015 1:07am

You can run a 2-node FCI and replicate in a standalone instance via Availability Group. The standalone instance needs to be a node in the WSFC that the FCI belongs to. You basically have a 3-node WSFC running a 2-node FCI and a standalone SQL Server instance. Automatic failover happens between the nodes of the FCI but not between the replicas of the AG. Refer to this whitepaper for more details.
Free Windows Admin Tool Kit Click here and download it now
August 27th, 2015 2:30pm

Hi, Edwin:

Would I interpret your response to be:

Cluster three servers

On two of them, install clustered instances of SQL, thereby creating the FCI.

On a third, clustered server, install an un-clustered instance, (even if the third server is also home to clustered instances and can participate in the FCI)

On the unclustered instance, create a primary AAG containing ASPStateInMemory.  Create the AAG as synchronous with automatic failover.

Designate two clustered instances as home to replicas of the AAG.

Why need I create an unclustered instance to contain the primary replica?  Why shouldn't it be a clustered instance, like all the others?

I wonder whether I can have clustered and unclustered instance of SQL on the same server.




August 28th, 2015 8:51am

And, I forgot to ask:

If I have turned on durability for an ASPStateInMemory database (hekaton), and placed it in an instance on clustered SQL Node1, when the instance fails to clustered SQL Node2, is state in the database preserved?

Free Windows Admin Tool Kit Click here and download it now
August 28th, 2015 8:57am

You only install a standalone instance on the third server, not join it to the FCI on the 2 servers. You can't have an AG where the secondary replica is on the same server as the primary replica. On the FCI, you create the ASPStateInMemory database so you have local high availability and automatic failover. For DR, you add it to an AG with the standalone instance as the secondary replica.

You can have a FCI and standalone instance on the same server. It's just not practical

August 28th, 2015 10:54am

Hi, Edwin:

Would I interpret your response to be:

Cluster three servers

On two of them, install clustered instances of SQL, thereby creating the FCI.

On a third, clustered server, install an un-clustered instance, (even if the third server is also home to clustered instances and can participate in the FCI)

On the unclustered instance, create a primary AAG containing ASPStateInMemory.  Create the AAG as synchronous with automatic failover.

Designate two clustered instances as home to replicas of the AAG.

Why need I create an unclustered instance to contain the primary replica?  Why shouldn't it be a clustered instance, like all the others?

I wonder whether I can have clustered and unclustered instance of SQL on the same server.




Free Windows Admin Tool Kit Click here and download it now
August 28th, 2015 12:50pm

Thanks for you help, Edwin.

Somehow, I had gotten the idea that once I had turned on 'durability' for ASPStateInMemory, and had placed the database on a SQL Instance that was a member of an FCI, that when the time came to move the instance that supported the ASPStateInMemory database from one node to another, that without a DAG, the on-disk state information would be lost.  I had thought there wa a magic reason the DAG was required.

It's nice to know that FCI is enough to support ASPState durability.

Thanks, Bob!

August 28th, 2015 5:59pm

This topic is archived. No further replies will be accepted.

Other recent topics Other recent topics