none
RDS High Availability - What should it act like? RRS feed

  • 問題

  • Hello,

    Setting up a lab, trying to make an RDS farm be "highly available"... but it doesn't seem right.  1st attempt, not sure what to expect (expecting too much?).  My main question is, can I setup RD Web, Gateway, Broker (server 2016) using MS tools to make them HA?  Ignore session host and VDI for now.  Any tips on what I may have done wrong so far?

    Scenario
    All server roles configured for HA (trying).  Say an external person logs into RD Web, connects to his office PC.  Can I reboot one of the servers in the HA pairs to do maintenance and not effect the RDP session?  I'm fine with a brief pause in service (IE: screen pauses for a second or two while it switches to another node/host).

    My Issue

    In my lab, when I reboot one of my web/gateway pairs, the RDP session to a PC gets disconnected and cannot reconnect.  HA isn't working it seems.  Or is that only for Session Host sessions in a RDS collection?

    My Lab Server Names/Roles
    Running on PC with Server2016 Hyper-V Host

    • HyperV Host Virtual Switch is private network for all VMs
    • All VMs running Server2016 (except Win10 VM)
    • AD (ADDS,DCHP,DNS)
    • RDWebGw-01 (RD web access, gateway, NLB unicast)
    • RDWebGw-02 (RD web access, gateway, NLB unicast)
    • RDB-01 (RD broker)
    • RDB-02 (RD broker)
    • RDSH-01 (RD session host)
    • RDSH-02 (RD session host)
    • Win10 PC (also running sql for broker)

    Another tidbit.  When the RDP session to the PC is active, I can see the active session in the RD Gateway Manager on only one of the two gateways.  I had thought I should be able to see it on both Gateways, but I was wrong (in my lab).  Seem right to you too?


    • 已編輯 Drew.Prince 2019年8月23日 上午 02:25 Add more info
    2019年8月23日 上午 02:05

解答

  • Q) My main question is, can I setup RD Web, Gateway, Broker (server 2016) using MS tools to make them HA?

    A) Yes, kind of.  And this is what occurs in my lab to give other people an idea of what may happen when various roles fail.  If I'm wrong I would like to learn more of how you have it setup (using MS tools).

    If anyone else finds this, I hope this can be of use to you.  My expectation of "Highly Available" was not correct.  I expected that someone could RDP to an internal desktop or session host with zero interruption, or very little (IE: 1-2 seconds of downtime while a failed system reverts to another working system).

    In my findings in my lab this is what happens when a role goes offline:

    1) Web Access (with NLB)
    When the WA role goes offline the MS NLB solution takes a few seconds to fail over, the 1st few seconds are the delay in NLB to wait for the heartbeat to fail then timeout, then a few more seconds to switch NLB hosts.  On a web server this is usually not an issue, and someone would have to hit the website during those few seconds to display an error.

    2) Gateway (with NLB)
    In my lab, RD connections appear on one of the two GW hosts (not both).  Then when one of the hosts fail, the RD session stops working, shows as disconnected.  In my lab it takes about 10-20 seconds for the RD client to reconnect to the other gateway server.

    3) Broker (with NLB and external SQL)
    In my lab, a broker server can fail and the RD client does not get interrupted at all which is great.

    4) Session Host
    If someone is connected to a session host and that host fails, yes their connection will drop and is expected.  However, if you have multiple session hosts they can log on again and continue working from the last point that they saved their work.

    • 已標示為解答 Drew.Prince 2019年9月9日 下午 02:16
    2019年9月9日 下午 02:13

所有回覆

  • Hi,

    >My main question is, can I setup RD Web, Gateway, Broker (server 2016) using MS tools to make them HA?
    It is supported to configured all RDS roles with HA. However, each component HA working principle is not same, 3rd party load balance configuration/tool can be considered to improve the HA function.

    >I can see the active session in the RD Gateway Manager on only one of the two gateways.  I had thought I should be able to see it on both Gateways.
    We may need more configuration to configure the load balance for Gateway/ such as NLB, hardware-based load balancing, or DNS round robin, in order to split request to both gateway servers.

    Below is an article which describes the details, can be considered as reference:
    https://www.microsoftpressstore.com/articles/article.aspx?p=2346349&seqNum=4

    Best Regards,
    Eve Wang

    Please remember to mark the replies as answers if they help.
    If you have feedback for TechNet Subscriber Support, contact tnmff@microsoft.com.

    2019年8月23日 上午 09:07
    版主
    • It is supported to configured all RDS roles with HA.

    Yes, I know that.

    • However, each component HA working principle is not same, 3rd party load balance configuration/tool can be considered to improve the HA function.

    Of course. 

    This did not answer any of my questions.  Should I be able to reboot servers and not affect RDP sessions to PCs?

    • We may need more configuration to configure the load balance for Gateway/ such as NLB, hardware-based load balancing, or DNS round robin, in order to split request to both gateway servers.

    Again, I don't think you read my comments very clearly.  Again, I'm using MS roles, including the MS NLB using unicast (not RRDNS)

    2019年8月23日 下午 03:56
  • Hi,

    >when I reboot one of my web/gateway pairs, the RDP session to a PC gets disconnected and cannot reconnect.

    >When the RDP session to the PC is active, I can see the active session in the RD Gateway Manager on only one of the two gateways.

    Based on your description, it seems that your NLB configuration is not working as expected, in general, it should direct connection to both nodes. If possible, please check Event Viewer and try to find relate event, it may provide helpful information about the problem.

    Besides, NLB is another technology which is beyond the support scope on RDS, I would recommend to you post on NLB forum – High Availability – have a confirmation of NLB deployment and make sure it is working as expected first:
    https://social.technet.microsoft.com/Forums/en-US/home?forum=winserverClustering

    Thank you for your understanding. 

    Best Regards,
    Eve Wang

    Please remember to mark the replies as answers if they help.
    If you have feedback for TechNet Subscriber Support, contact tnmff@microsoft.com.


    2019年8月26日 上午 02:46
    版主
  • Hi,

    How things are going there on this issue?

    Please let me know if you would like further assistance.

    Best Regards,
    Eve Wang 

    Please remember to mark the replies as answers if they help.
    If you have feedback for TechNet Subscriber Support, contact tnmff@microsoft.com.

    2019年8月27日 上午 09:23
    版主
  • Hi,

    Is there any update?

    Best Regards,
    Eve Wang

    Please remember to mark the replies as answers if they help.
    If you have feedback for TechNet Subscriber Support, contact tnmff@microsoft.com.

    2019年8月29日 上午 06:49
    版主
  • I am still waiting for someone to answer my question.
    2019年8月29日 下午 02:03
  • Hi,

    Please confirm the detail configuration steps via “Configuring a Windows 2012 Server Remote Desktop Gateway NLB cluster”:
    https://blogs.technet.microsoft.com/marobbin2/2012/10/10/configuring-a-windows-2012-server-remote-desktop-gateway-nlb-cluster/

    Also, please shutdown one of the Gateway, then, if problem happens, please check event on both client and Gateway and confirm that if there is any relate event has been logged:
    Event Viewer -> Applications and Services Logs -> Microsoft -> Windows -> TerminalServices and RemoteDesktopServices

    Best Regards,
    Eve Wang

    Please remember to mark the replies as answers if they help.
    If you have feedback for TechNet Subscriber Support, contact tnmff@microsoft.com.

    2019年8月30日 上午 03:29
    版主
  • Q) My main question is, can I setup RD Web, Gateway, Broker (server 2016) using MS tools to make them HA?

    A) Yes, kind of.  And this is what occurs in my lab to give other people an idea of what may happen when various roles fail.  If I'm wrong I would like to learn more of how you have it setup (using MS tools).

    If anyone else finds this, I hope this can be of use to you.  My expectation of "Highly Available" was not correct.  I expected that someone could RDP to an internal desktop or session host with zero interruption, or very little (IE: 1-2 seconds of downtime while a failed system reverts to another working system).

    In my findings in my lab this is what happens when a role goes offline:

    1) Web Access (with NLB)
    When the WA role goes offline the MS NLB solution takes a few seconds to fail over, the 1st few seconds are the delay in NLB to wait for the heartbeat to fail then timeout, then a few more seconds to switch NLB hosts.  On a web server this is usually not an issue, and someone would have to hit the website during those few seconds to display an error.

    2) Gateway (with NLB)
    In my lab, RD connections appear on one of the two GW hosts (not both).  Then when one of the hosts fail, the RD session stops working, shows as disconnected.  In my lab it takes about 10-20 seconds for the RD client to reconnect to the other gateway server.

    3) Broker (with NLB and external SQL)
    In my lab, a broker server can fail and the RD client does not get interrupted at all which is great.

    4) Session Host
    If someone is connected to a session host and that host fails, yes their connection will drop and is expected.  However, if you have multiple session hosts they can log on again and continue working from the last point that they saved their work.

    • 已標示為解答 Drew.Prince 2019年9月9日 下午 02:16
    2019年9月9日 下午 02:13