2014 Always on multi subnet custer

  • Hello,

    I'm looking to setup a multi subnet cluster between two DC's using a MPLS which is very reliable. There will be 3 nodes in each DC. My question is related to the quorum settings that I should implement. In one DC to break the possible even quorum vote I've added a file share witness.

    The difference with this setup however is that if the MPLS goes down and quorum is lost between DC's I do not want the DC without the majority vote to go offline if and when it's separated from the DC with majority vote. This is not a DR site. We will have live traffic in the DC. I understand that with Always on the secondary replica will fall behind as it can't be updated but we would rather have this and have the DB's accessible. If we determine the MPLS will be down for an extended period of time we can then take corrective action by either moving traffic or removing the DB from always on etc.

    Essentially I'm asking how do I ensure split brain continues for a multi subnet failure?

    I've already increased the CrossSubnetThreshold and duration for the cluster to ensure the heartbeat across the WAN isn't as sensitive. This should presumably cover shorter outages but the question would be how to ensure the secondary site remains up.

    Thanks

  • You don't mention what version of Windows you are using, but regardless you can adjust the quorum vote that each server gets to ensure that you maintain qurorum on your primary site in the event the remote one goes down.

    I normally do this using Powershell (as it is easier to manage).

    To find the current nodeweight of each server, from an admin shell on a machine within the cluster (for simplicity)

    get-clusternode | ft Name, NodeWeight

    This will list all of your servers, and whether they have weight (ie a quorum vote).

    You can remove a quorum vote from a server with

    (Get-ClusterNode <Server Name>).NodeWeight = 0

    Adding back again you just set the NodeWeight to 1.

    If running 2012 R2 then I would also recommend looking at dynamic quorum, so that you can utilize extra resources to dynamically ensure that you maintain quorum in your primary site.



    Shamless self promotion - read my blog http://sirsql.net

  • Thanks, apologies for not providing this info earlier. OS version is Server 2012 R2

    So for adjusting the vote - let's say i have 4 votes in primary site (each of the 3 nodes and a file share) and 3 in the secondary (each of the 3 nodes)

    The primary site should always in theory have the majority and remain up as the file share is on that DC. My question is what happens to the 3 nodes in the Secondary? How do I ensure they remain up when the WAN link goes down?

    I do not want these to shut down etc or be evicted. There are other Always on AG's which are running solely in that DC between nodes. The reason for the cluster to span the WAN and all nodes is a business requirement for 1 single DB to be present on all nodes in the cluster. Always on obviously makes it easy to have read-only copies of the DB on all nodes.

    Should I in this case configure it so that split brain does occur? 3 votes each side?

    Not sure if the image will display but I require the DB3 AG to remain up on the secondary even if DB1 is disconnected from the primary DC

Viewing 3 posts - 1 through 2 (of 2 total)

You must be logged in to reply to this topic. Login to reply