Wednesday, June 05, 2019

Actual Emails: Will MSOLEDB work for connecting to an Availability Group?

Scenario:

We need to configure an existing legacy application from an external vendor to talk to our new SQL Server 2017 Availability Group, which spans multiple subnets. In the end, that last bit is the key. The old data provider MSOLEDB will work for connecting to single-subnet Availability Group listeners, but won't work consistently when connecting to a multisubnet Availability Group's listener. The key is the ability to specify MultiSubNetFailover=True in the connection parameters.

Client's software vendor:
The connection used is the Microsoft OLE DB Provider for SQL Server that is supplied by Microsoft to create the Data Link Properties. When configuring our Data Link, we use the "Microsoft OLE DB Provider for SQL Server". The connection string is formatted: 
Provider=SQLOLEDB.1;Password="whatever";Persist Security Info=True;User ID=username;Initial Catalog=Test;Data Source=ServerName 
Answer:
Good info, but we do need to make a change here. SQLOLEDB is the provider from back in the SQL 2000 era. Do not recommend its use for new development. It has been replaced by the Native Client (SNAC), which has since been replaced by MSOLEDBSQL (I linked below).  It should be easy and transparent to upgrade the provider from SQLOLEDB with no negative impact.
Here’s why we need to upgrade the data provider to talk to our SQL Availability Group. The SQL Server Listener for a multi-subnet Availability Group actually has two IP’s. When you perform at a command line:
Nslookup SQLListenerNameWhatever 
You get back an IP in each subnet (in our case, two), for each replica SQL instance in the Availability Group. 
When a connection string uses MultiSubNetFailover=True and connecting to the Availability Group Listener name (not the IP or either SQL Server instance name), BOTH IP’s are tried simultaneously and immediately, and the driver talks only to the IP that replies: the primary replica.  After a failover, the other IP begins to reply immediately, and so there is no delay in reconnectivity when a failover occurs. 
Without specifying MultiSubNetFailover=True, your application will (essentially randomly) pick ONE of the two IP’s for the Listener, and try it. There is no way to “rig” one IP to be first consistently over time. If it picks the primary replica, everything works! If it picks the IP for the current secondary replica… your application’s connection timeout will have to expire and then try the next IP.  This is why I’m bringing this up – the application will timeout upon SQL login without MultiSubNetFailover=True.
This hasn’t been an issue with your other clients if they aren’t using a multisubnet availability group. If they have an Availability Group all inside only one subnet, then the Listener only has one IP in DNS, and MultiSubNetFailover=True isn’t required.
You should be fine to install the MSOLEDBSQL provider released in 2018 and use that in your data link. Obviously it should be tested for due diligence, but it should work. At the very least, you could try instead the SQL Native Client 11 (SQLNCLI11), which was released for SQL Server 2012, and it also should work just fine for both OLEDB or ODBC.
Let me know if you have any more questions.
Connection string information for SQL Server Availability Groups:
  • In SSIS, the new MSOLEDBSQL OLE DB driver appears as "Native OLE DB\Microsoft OLE DB Driver for SQL Server". The old OLE DB driver is "Native OLE DB\Microsoft OLE DB Provider for SQL Server". 
Note: you also need to make sure your Windows Cluster has RegisterAllProviderIP's set to ON for a multisubnet Availability Group!

Edit March 2023: Added latest OLE DB/ODBC information.