We are testing AlwaysOn Availability Groups on a WSFC multi-subnet cluster and using dynamic DNS registration. Our issue is that sometimes the IP for the Availability Group Listener (AGL) in the active subnet is registered in DNS as well as the IP for the AGL that is in the inactive subnet.
When we initially setup the cluster, by default, RegisterAllProvidersIP is set to 1 so we expect both the active IP and inactive IP of the AGL to be registered in DNS. We followed the steps outlined by the MS AlwaysOn Pro team to change RegisterAllProvidersIP to 0 so that only the IP in the active site is registered with DNS. We performed extensive testing once we made this change and discovered the following behavior:
- About70% of the time, only the active site’s IP for the AGL is registered in DNS (Good!)
- About 30% of the time, both the active IP and inactive IP for the AGL are registered in DNS (Bad!)
There is no pattern to the issue. We’ve looked at DNS when multiple records are created and they both have the exact same date/time stamps so the records are obviously being created at the same time. To clean up this issue, we simply run the following command and only the active IP is registered:
Get-ClusterResource AGL_Network_Name | Update-ClusterNetworkNameResource
Is anyone successfully using multi-subnet clustering with AA AGs in a production environment? If so, I’d appreciate any help you might be able to provide.
We’ve confirmed the issue occurs running under VMware and with physical servers. The VMs and physicals are located in physically different datacenters and subnets. We have a fast pipe between them and latency is minimal. The VMs are running Server 2012 and the physicals Server 2012 R2. SQL Server 2012 SP2 is being used.
Best Answer
This document from Microsoft lists a known issue with .net 3.5 erroring out at times. While not specifically identifying your issue in DNS --It may be the cause and is probably worth reading:
https://msdn.microsoft.com/en-us/library/hh213080.aspx#MultiSubnetFailover