We’ll be upgrading our SQL cluster masters to SQL 2016 CU1 on Wednesday night at 9:00PM EDT (01:00 August 4th UTC). The outage window should be minimal, a few minutes at most as we move all services to the current replicas.
Details
One of the unexpected issues we hit with the SQL 2016 upgrade was a performance degradation, resulting in much higher CPU usage on the masters of our SQL cluster with 354 databases. We narrowed this down to an issue with the AlwaysOn Availability Group DMVs we use to get status for things like Opserver, but also use in the backup scripts we use by Ola Hallengren.
Specifically, the
sys.dm_hadr_availability_replica_states view became very expensive in certain scenarios such as the high database counts we have on that cluster. We worked with Microsoft on what became KB3173038 to identify and test fixes for the issue - fixes which we could not apply in a long-term way without a proper hotfix, cumulative update, or service pack (it needs a resource database change). This fix is included in SQL 2016 CU1 though, so we’ll be applying that upgrade on Wednesday night. It is already live on our development tier and replicas.
For updates during the upgrade, follow @StackStatus on Twitter.