none
2 Node DAG - Unexpected Failover Behaviour RRS feed

  • Question

  • Hi all, please consider this scenario...

    AD Site A contains Exchange Server A and also the witness server for the DAG.

    AD Site B contains Exchange Server B.

    All DBs are mounted on Exchange Server A.

    I powered off the witness server, exchange remains functional on Server A as expected.

    I then power off Exchange Server A. My expectation here is that DBs would remain dismounted as there is no witness server to allow Exchange Server B the majority votes required to take control of the DAG.

    However, what actually happened was Exchange Server B just mounted its passive copies of the DBs and exchange remained functional. How is this possible? What am I missing here? I thought with no witness server and a failure of your primary node, you would be dead in the water?

    Thanks in advance.

    Wednesday, July 17, 2019 1:30 PM

Answers

  • It sure does sound like it reading that, however, I did find something further only minutes after posting this!

    Basically there was a bogus DNS record in my environment. So whereas I "thought" I was shutting down the witness server, a DNS record was pointing that server name to a completely different IP address. So after I powered it off, exchange could still see something running on that IP and was therefore using it.

    I just cleaned up that old DNS record and re-ran my test. Sure enough I could see that after shutting down the witness server and my primary node, exchange is now dead. I also spotted my cluster quorum parameter was set to majority and not node and file share witness so corrected that too.

    • Marked as answer by RecLambyUK Wednesday, July 17, 2019 2:33 PM
    Wednesday, July 17, 2019 2:33 PM

All replies

  • Sounds like Dynamic Quorum is doing it's job:

    https://practical365.com/exchange-server/windows-server-2012-dynamic-quorum/


    My Blog: http://exchangeitup.net My Twitter: http://twitter.com/ExchangeITup

    Wednesday, July 17, 2019 2:06 PM
  • It sure does sound like it reading that, however, I did find something further only minutes after posting this!

    Basically there was a bogus DNS record in my environment. So whereas I "thought" I was shutting down the witness server, a DNS record was pointing that server name to a completely different IP address. So after I powered it off, exchange could still see something running on that IP and was therefore using it.

    I just cleaned up that old DNS record and re-ran my test. Sure enough I could see that after shutting down the witness server and my primary node, exchange is now dead. I also spotted my cluster quorum parameter was set to majority and not node and file share witness so corrected that too.

    • Marked as answer by RecLambyUK Wednesday, July 17, 2019 2:33 PM
    Wednesday, July 17, 2019 2:33 PM
  • Also, in my case the OS is server 2016 so not sure dynamic quorum even comes into play does it?
    Wednesday, July 17, 2019 2:38 PM