I keep getting questions on what is improved other than the increased number of secondaries(8 now) with AlwaysOn Availability Group(AG) in SQL 2014. In this blog post  I would like to talk about one major enhancement around the increased readable secondaries availability in SQL Server 2014.

In SQL Server 2012 there is a good chance that your read workloads can be killed during network failures. Typically this happens in a multi subnet environment where you have AG nodes in 2 or more data centers.If network failure results a communication failure between primary and secondary replicas, then the secondaries will go into resolving state.In resolving state, the read workloads are basically killed. (Yes, yet another reason why you need to test your AG deployments)

With SQL Server 2014, there is good news. Read workloads are not impacted during network failures, primary replica down or cluster quorum issues.

Yes, it stays up all the time, and will be serving your read workloads. The replica will be in resolving state, however the databases are still available for your read connections/workloads.

Keep in mind : Read-only routing is not yet supported for this, and you cannot use listener for your reporting workloads.Brent Ozar calls out this in his blog post here.

Lets now walk though this enhancement :

I will be using a 2 node AG setup and the data transfer mode(Availability Mode) is asynchronous.

If I query the DMV sys.dm_hadr_availability_replica_states I can see the replica states.

SELECT role_desc,connected_state_desc  FROM sys.dm_hadr_availability_replica_states

role_desc               connected_state_desc

PRIMARY              CONNECTED

SECONDARY        CONNECTED

Lets double-check if the data transfer is set as asynchronous

SELECT  is_primary_replica, synchronization_state_desc FROM sys.dm_hadr_database_replica_states

is_primary_replica       synchronization_state_desc

0                                  SYNCHRONIZING

1                                  SYNCHRONIZED

Everything looks good, and now lets take down the primary replica.

Note : Dont try this at home ! (I mean production)

Okay, As this is a test instance I really don’t care and what I did was shutdown the SQL Services to take the primary down.

If we query the DMV again, we can see

SELECT role_desc,connected_state_desc  FROM sys.dm_hadr_availability_replica_states

role_desc                connected_state_desc

RESOLVING           DISCONNECTED

And note, there is no entry for primary as we I have taken it down.

Even though the secondary replica is in resolving state, read connections just works.

USE DemoAG
GO
SELECT COUNT (*) AS [TotalRowCount] FROM [dbo].[InsertCount]

TotalRowCount

6886

The secondary database is in NOT SYNCHRONIZING, NOT_HEALTHY state, however its ONLINE.

SELECT synchronization_state_desc,synchronization_health_desc,database_state_desc FROM sys.dm_hadr_database_replica_states

synchronization_state_desc         synchronization_health_desc       database_state_desc

NOT SYNCHRONIZING               NOT_HEALTHY                           ONLINE

Here is a view from Management Studio :

Conclusion:

Yes, this is a very good enhancement which will make sure your readable secondaries are highly available and I’m pretty sure that this is one of those enhancement which will force you to deploy AGs in SQL 2014 rather than opting SQL 2012.

Thanks for reading and keep watching this space for more !