Multi-Site Cluster SQL Server 2016 AlwaysON Infrastructure

  • Hello,
    I have two production site. I want to design a new SQL Server 2016 AlwaysOn Cluster for two production sites. I need to have High Availability at each site because I have mission critical databases at each site. Also I need to a Disaster Recovery for each site.
    I can design two different clusters at their site and one DR node across site like below. So I need to have three nodes for each site(total 6)


    I thought maybe i can combine two cluster in one cluster like below. As you can see below, each local nodes have synchronous replication and asynchronous replication databases. Four nodes are looking enough for this scenario.
    I have not tested this design. So I have some questions like below;
    1)      Cluster also go down, If wan line will be lost between two sites.
    2)      Can each nodes works fine at their local site, If wan line will be lost between two sites.
    3)      Do I need to have File Share Witness to avoid cluster go down?
    4)      If I need to have a witness, which location is the best for location (btw, I have third site location)
    5)      What do you think about this design? Is there any risk which I don’t see?
    6)      Is maintenance operation easy or not?

    Thank you.

  • myalliku - Thursday, March 30, 2017 2:28 PM

    Hello,
    I have two production site. I want to design a new SQL Server 2016 AlwaysOn Cluster for two production sites. I need to have High Availability at each site because I have mission critical databases at each site. Also I need to a Disaster Recovery for each site.
    I can design two different clusters at their site and one DR node across site like below. So I need to have three nodes for each site(total 6)


    I thought maybe i can combine two cluster in one cluster like below. As you can see below, each local nodes have synchronous replication and asynchronous replication databases. Four nodes are looking enough for this scenario.
    I have not tested this design. So I have some questions like below;
    1)      Cluster also go down, If wan line will be lost between two sites.
    2)      Can each nodes works fine at their local site, If wan line will be lost between two sites.
    3)      Do I need to have File Share Witness to avoid cluster go down?
    4)      If I need to have a witness, which location is the best for location (btw, I have third site location)
    5)      What do you think about this design? Is there any risk which I don’t see?
    6)      Is maintenance operation easy or not?

    Thank you.

    Some important information is missing from this mock-up. What are the business requirements? Do you need to service applications in each site separately from the other site or is one site truly a DR site for the other site?

    I have plenty of multi-site single cluster setups. I also have AG enabled on these multi-site clusters. If one site is truly a DR, you need to consider the fileshare witness and a fileshare witness should be available at each site (true DR setup).

    Jason...AKA CirqueDeSQLeil
    _______________________________________________
    I have given a name to my pain...MCM SQL Server, MVP
    SQL RNNR
    Posting Performance Based Questions - Gail Shaw[/url]
    Learn Extended Events

  • Our each sites have independent applications from other sites. So, if an application is storing at Site1, database also stroring at Site1. Network latency or network interrupt is critical. If I store all databases at one site, when wan line lost, all applications will be stop at cross site. Thats why, I need to design an infrastructure which works as active at each sites.

    When I investigate 4 nodes multi-cluster site design again, I think it is not robust-ness. Maybe I can use Cloud Witness or Fileshare on third site, but I can lost these witness more possibility than a witness in local. In the first picture, Node1, Node2 and Witness in same site, DR in other Site. If I lost Site1, Cluster would stop, because there is no Witness and other nodes. But I think that I can start cluster by force command on DR site (net start clussvc /fq )

    What do you think about design which I share first? Two nodes on site for HA and one node for DR. If you have a diagram for multi-site cluster, can you share please?

    Regards.

  • myalliku - Thursday, March 30, 2017 2:28 PM

    I thought maybe i can combine two cluster in one cluster like below. As you can see below, each local nodes have synchronous replication and asynchronous replication databases. Four nodes are looking enough for this scenario.
    I have not tested this design. So I have some questions like below;
    1)      Cluster also go down, If wan line will be lost between two sites.
    2)      Can each nodes works fine at their local site, If wan line will be lost between two sites.
    3)      Do I need to have File Share Witness to avoid cluster go down?
    4)      If I need to have a witness, which location is the best for location (btw, I have third site location)
    5)      What do you think about this design? Is there any risk which I don’t see?
    6)      Is maintenance operation easy or not?

    Your diagrams are good but it's still not clear what you have or plan to have. If all 6 nodes are part of the same cluster you will have to do something about the witness votes. It looks like each site is a DR for the other, is that correct?

    -----------------------------------------------------------------------------------------------------------

    "Ya can't make an omelette without breaking just a few eggs" 😉

  • Hi Perry,

    Somewhere up in the above threads you mention a FSW at each site? Could you elaborate on that a little bit please? I thought an ideal FSW  is located at a third location?

    Cheers
    Alex

  • alex.sqldba - Tuesday, April 11, 2017 1:27 AM

    Hi Perry,

    Somewhere up in the above threads you mention a FSW at each site? Could you elaborate on that a little bit please? I thought an ideal FSW  is located at a third location?

    Cheers
    Alex

    where?

    -----------------------------------------------------------------------------------------------------------

    "Ya can't make an omelette without breaking just a few eggs" 😉

  • "If one site is truly a DR, you need to consider the fileshare witness and a fileshare witness should be available at each site (true DR setup)."

    Er.. Sorry quoting feature is not working properly in Edge.... Seemed to have stopped working yesterday! So had to copy and paste.

    It's the part about the fileshare witness being available at each site that is making me think you are suggesting a FSW are Location1 and Location2.

    Cheers
    Alex

  • This is good advices for this problem

  • Oh. I think I have been being an idiot. A chronic condition.
    I'd not realised there are two independent clusters being discussed.
    I thought the illustrations were depicting a 4 node, but with a logical divisions for availability group.

    I am clustered out this week. Where is Easter?!

  • Perry Whittle - Friday, March 31, 2017 11:13 AM

    myalliku - Thursday, March 30, 2017 2:28 PM

    I thought maybe i can combine two cluster in one cluster like below. As you can see below, each local nodes have synchronous replication and asynchronous replication databases. Four nodes are looking enough for this scenario.
    I have not tested this design. So I have some questions like below;
    1)      Cluster also go down, If wan line will be lost between two sites.
    2)      Can each nodes works fine at their local site, If wan line will be lost between two sites.
    3)      Do I need to have File Share Witness to avoid cluster go down?
    4)      If I need to have a witness, which location is the best for location (btw, I have third site location)
    5)      What do you think about this design? Is there any risk which I don’t see?
    6)      Is maintenance operation easy or not?

    Your diagrams are good but it's still not clear what you have or plan to have. If all 6 nodes are part of the same cluster you will have to do something about the witness votes. It looks like each site is a DR for the other, is that correct?

    Hey Perry,
    I will not apply third picture diagram which has 4 nodes in two different sites in one cluster, because It is not looking robust.
    So, I decided to apply first and second diagrams seperately. I think that this method(two node for HA + one node for DR) is best practice.

    Thank you for your advice.

Viewing 10 posts - 1 through 9 (of 9 total)

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