none
Availability Groups and Cluster Quorum

    Question

  • Hello.

    I've set up a two node cluster on the same subnet, SQLDEVCLUST to test Availability Groups. As this is AG I’m not using shared storage. 

    It is made up of two SQL Servers, SQLDEV3 & SQLDEV4.

    I added a file share to the cluster and set the Quorum Config to Node and File Share Majority.

    Cluster manager appears happy with this set up.

    I successfully configured a AG between SQLDEV3 (primary) & SQLDEV 4 (secondary).

    I decided to do a manual failover using the wizard. In the Failover Readiness box reports this: The current WSFC cluster quorum vote configuration is not recommended for this availability group.

    I ignored this (just so I could see what it said on the summary screen).  It reported: The replica is on a WSFC node that does not have a quorum vote.

    But according to BOL:

    By default, each node in the WSFC cluster is included as a member of the cluster quorum; each node has a single vote in determining the overall cluster health, and each node will continuously attempt to establish a quorum.

    Also in View Cluster Quorum Information on the AG dashboard the vote count for both SQLDEV3 & SQLDEV4 reports "Not available".

    Anyway I decided to proceed and manually failed over with no problem apart from a warning about the quorum config again.

    I stopped the SQLServer service on SQLDEV3 to simulate a failover and it failed over successfully.

    I switched SQLDEV3 off completely, failover was again successful.

    Disabled the network cards, failover successful

    Basically failover appears to be working fine but SSMS/AG doesn’t like my config?

    I’m new to clustering so not quite sure what I may have got wrong here.

    Thanks

    Friday, May 18, 2012 2:08 PM

Answers

  • Hey Rob,

    Have you set (or do you know if someone has set) quorum node weight?

    You could first try changing quorum model and back again to see if that helps but if not you should read my blog post on the subject of node weight here: Weight doesn’t ALWAYS have to be AlwaysOn

    Essentially once you have the hotfix in place you should be able to query and set the nodeweight property. If you find any nodes are zero then set them to one -if that is what you require.

    Regards,
    Mark Broadbent.

    Contact me through (twitter|blog)

    Please click "Propose As Answer" if a post solves your problem
    or "Vote As Helpful" if a post has been useful to you

    • Marked as answer by SQL Rob Monday, May 21, 2012 12:53 PM
    Friday, May 18, 2012 6:53 PM

All replies

  • Hey Rob,

    Have you set (or do you know if someone has set) quorum node weight?

    You could first try changing quorum model and back again to see if that helps but if not you should read my blog post on the subject of node weight here: Weight doesn’t ALWAYS have to be AlwaysOn

    Essentially once you have the hotfix in place you should be able to query and set the nodeweight property. If you find any nodes are zero then set them to one -if that is what you require.

    Regards,
    Mark Broadbent.

    Contact me through (twitter|blog)

    Please click "Propose As Answer" if a post solves your problem
    or "Vote As Helpful" if a post has been useful to you

    • Marked as answer by SQL Rob Monday, May 21, 2012 12:53 PM
    Friday, May 18, 2012 6:53 PM
  • If you added the file share to the same cluster as sqldev3 and sqldev4 then this is not supported. The File Share MUST be on a separate server (or WSFC to make the File Share highly available). The File Share is a Witness that counts for a single vote in a dispute over which node(s) have ownership over resources.

    So, to correct the problem, create a File Share Witness on another server, give it the appropriate permissions, then change your quorum configuration to use the (remote) FSW.

    Monday, May 21, 2012 5:58 AM
  • Thanks for the replies.

    File Share: this is already on a separate server, it would have obviously helped if I'd mentioned that in my original post.

    Node Weight: I or anyone else hadn't altered the setting. I installed the hot fix and all the node weights = 1. I altered one of the nodes to 0 and then back to 1. Went back into SSMS and and the AG dashboard was now happy with the quorum config. In the View Cluster Quorum Information screen the vote count is now 1 where it was previously "not available". Don't understand why doing that would fix it, but it has. Very useful info in that blogpost btw. 

    cheers

     

    • Marked as answer by SQL Rob Monday, May 21, 2012 9:37 AM
    • Unmarked as answer by SQL Rob Monday, May 21, 2012 12:52 PM
    Monday, May 21, 2012 9:37 AM
  • Thanks for feedback on my node weight post. Not so useful as to deserve a Marked Answer or Vote up though? :)

    Regards,
    Mark Broadbent.

    Contact me through (twitter|blog)

    Please click "Propose As Answer" if a post solves your problem
    or "Vote As Helpful" if a post has been useful to you

    • Marked as answer by SQL Rob Monday, May 21, 2012 12:53 PM
    • Unmarked as answer by SQL Rob Monday, May 21, 2012 12:53 PM
    Monday, May 21, 2012 9:54 AM
  • It is not very helpful that Microsoft in its description of availability groups setup says the following:

    In contrast to database mirroring, there is no witness role in Always On Availability Groups.

    I copied that statement directly from MSDN. It creates a lot of confusion when in one breath "there is no witness" and then you find out that you need a file share to "act as a witness".

    Common sense would tell you that you must have a file share in an availability groups configuration. Where else would you put backups? You would include the file share as a cluster resource to make it highly available.  It would be helpful to know what is required or availability groups will not work as opposed to what is best practice. Are these the same thing? I have seen videos on the net where availability groups are successfully set up and they do not even have a file share.

    Monday, October 22, 2012 2:23 PM
  • Hi Manitoba,

    I totally disagree with you. Why is that statement "not very helpful" when it is accurate. You are misunderstanding AlwaysOn AG thinking that the file share is the same thing as witness. AlwaysOn AG uses Windows Server Failover Clustering because (in part) it has adopted the Quorum mechanism to prevent and avoid split brain for exactly the same reason that Mirroring can use a Witness. File Share is not a requirement but *might* be needed depending upon the quorum note vote total.

    btw the file share is not used for backups but a quorum vote. You are misunderstanding availability groups configuration since your AG SQL instances will more times than not be local standalone instances configured within a WSFC.

    All the information is out there on MSDN, you just need to spend some time reading up on the subject.


    Regards,
    Mark Broadbent.

    Contact me through (twitter|blog|SQLCloud)

    Please click "Propose As Answer" if a post solves your problem
    or "Vote As Helpful" if a post has been useful to you
    Come and see me at the PASS Summit 2012

    Monday, October 22, 2012 3:06 PM
  • You are right about needing more time and what is really needed is to give it a try in a test mode.

    Thanks for your reply.

    Monday, October 22, 2012 3:17 PM