We are paying a hosting company to host our production environments. They have setup the entire sql server 2016 Enterprise Edition AlwaysOn architecture for us. I apologize if I misuse any of the always on terms as I am new to AlwaysOn and have just used Clusters previously. We have a .Net website running on 2 web servers with a load balancer, a SQL Server back-end database running on sql server 2016. The sql server 2016 is configured for always on in an availability group. With 2 synchronous replicas in one data center and and an asynchronous replica in another data center on the other side of the country basically. All things web related are VMs, all things SQL are dedicated physical boxes (at the recommendation of our hosting provider).
My question/possible issue is: when sql fails over automatically to the other synchronous server I sometimes experience (about 1/3 of the time) sql errors in the web application. Mostly they are errors like "could not complete the request" or "database x is participating in an availability group and is not accessible for queries". After a few seconds I can resubmit my request and the web site behaves normally. Is this the expected behavior during a fail-over? I was expecting slowness/lag but no error messages in the web application to the end user. Again this is just fail over between the synchronous replicas in my group by simply rebooting the primary replica from the command line (shutdown -r) , then alt+tab over to a web browser immediately and clicking around in the web site (aka- 1 user, which is me and not a high volume of transaction being thrown at the db). I have not attempted failing over to the asynchronous replica yet.
Is this expected behavior? I was under the impression I should see slowness during fail-over but no failed transactions/errors to the end user. I know we can make the secondary replica readable but we are not currently licensed for that I am told by the hosting provider. At the price we are paying for the hardware and licensing I am having a hard time believing this what we should be getting, especially during my single user test. I just want to make sure we have the best possible system setup before we go live. Thanks!