none
planned network outage and "response to resource failure"

    Question

  • Hi,

    We have a planned network outage on our hyper-v cluster. (The reason why is not important, the network team has assured us there's no way around it. Anyway...)

    In vmware, we have the "host isolation mode" setting, which allows us to tell the hosts to just leave the VM's as is and not try to fail them over or do anything other than just leave the vm's powered on and disconnected from the network along with the host.

    There is no "host isolation mode" on hyper-v, but i'm wondering if i can tweak the failover cluster settings like "response to resource failure" (on the cluster, not on each VM - will this set it for all VMs?) or failover tab under resource properties for each VM. Is it possible to achieve the same effect in hyper-v - i.e. when the host is completely disconected from all ethernet networks to just leave the VM's running and not try to restart or fail them over to another node?

    Ref. pictures


    This posting is provided "AS IS" with no warranties or guarantees and confers no rights


    • Edited by i686 Tuesday, April 04, 2017 12:04 PM
    Tuesday, April 04, 2017 12:03 PM

Answers

  • The biggest issue is not access to the VMs.  The issue you are going to run into is that the cluster nodes are going to lose access to each other.  As soon as a node no longer has access to the other nodes of the cluster the node will lose quorum and remove itself from the cluster. Any resources on that node will shut down in order to ensure no data is corrupted. There were changes made in 2016 to handle network outages for VMs to be able to handle this very scenario, but it is not available prior to 2016.

    tim

    Wednesday, April 05, 2017 1:31 PM

All replies

  • Hi,

    in event of planned outage on Hyper-V cluster node you have option to put Cluster node into Maintenance mode (there is option of stop cluster node via failover cluster manager). You can also drain roles from the cluster node so after this your Hyper-V cluster node will be isolated and no VMs or other roles will be running on this node. Also this node will be prevented to run VMs in event of failover on another node in cluster (if you have 3 ore more nodes in cluster).

    After you are done with maintenance on node, you can "resume" cluster node and fail back roles...

    Radek

    Tuesday, April 04, 2017 12:52 PM
  • Hi,

    As i mentioned, we have a planned network outage on our hyper-v cluster, not on one single host.


    This posting is provided "AS IS" with no warranties or guarantees and confers no rights

    Tuesday, April 04, 2017 1:44 PM
  • So you actually need your roles to be clustered but you want to prevent clustered roles to fail over? This is nonsense.

    You can just remove clustered roles (VMs) from cluster  and leave theme on their nodes just using clustered shared volumes...

    Radek

    Wednesday, April 05, 2017 6:00 AM
  • Hi,

    I don't mean to be rude, but please read my questions before answering. This is a temporary planned outage. The question is regarding "response to resource failure", not node failover or guest removal. I'm just trying to help a client, i'm not here to argue over personal opinions about what should and should not be done.


    This posting is provided "AS IS" with no warranties or guarantees and confers no rights


    • Edited by i686 Wednesday, April 05, 2017 6:53 AM
    Wednesday, April 05, 2017 6:22 AM
  • Hi Sir,

    I have tested it in my lab .

    (CSV disks are iSCSI disks, iscsi NIC not used for cluster/client )

    After the network outage , only the coordinate node can access the CSV disk .

    Based on this , the "CSV redirected access" mode doesn't work . It mean only HA VMs running on  CSV disk owner node may run (others will be put into "running-critical" mode).

    Hope it is helpful to you.

    Best Regards,

    Elton


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

    Wednesday, April 05, 2017 7:23 AM
    Moderator
  • Hi,

    you can prevent failover on individual VM by selecting "If resource fails, do not restart" .. (your first screen)

    You can also select node affinity (possible owners) by selecting advanced Policies tab. There you can specify on which node can individual VM actaully run in event of failover.

    You can also configure reaction of individual resource to network failures via Hyper-V manager / VM network adapter settings / advanced features / Protected network

    But I personally would go with removing resources from the cluster and after planned outage is done you can add theme to cluster again.

    Radek

    Wednesday, April 05, 2017 7:33 AM
  • Hi,

    Thank you, this was helpful.


    This posting is provided "AS IS" with no warranties or guarantees and confers no rights

    Wednesday, April 05, 2017 8:31 AM
  • Hi,

    Thank you. I should of course have mentioned our CSV is on FC, not iSCSI. So the storage will not be affected.


    This posting is provided "AS IS" with no warranties or guarantees and confers no rights

    Wednesday, April 05, 2017 8:32 AM
  • Hi,

    Thank you. I should of course have mentioned our CSV is on FC, not iSCSI. So the storage will not be affected.


    This posting is provided "AS IS" with no warranties or guarantees and confers no rights

    Hi Sir,

    I didn't cut the connection of iSCSI network only clusters .

    Even though it is FC , the disk should only owned by one cluster node at a time .

    If it is possible , you may test it in your lab   .

    Best Regards,

    Elton


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

    Wednesday, April 05, 2017 8:40 AM
    Moderator
  • The biggest issue is not access to the VMs.  The issue you are going to run into is that the cluster nodes are going to lose access to each other.  As soon as a node no longer has access to the other nodes of the cluster the node will lose quorum and remove itself from the cluster. Any resources on that node will shut down in order to ensure no data is corrupted. There were changes made in 2016 to handle network outages for VMs to be able to handle this very scenario, but it is not available prior to 2016.

    tim

    Wednesday, April 05, 2017 1:31 PM
  • Hi,

    Thanks. Ok, so what you're saying is that even with the "response to resource failure" set to do not restart, the nodes will all be removed from the cluster and the vms wil be powered off.

    So, i take it there is no workaround for 2012 R2 based on your answer. (todo: upgrade to 2016 ;-) )


    This posting is provided "AS IS" with no warranties or guarantees and confers no rights

    Thursday, April 06, 2017 8:17 AM
  • Hi,

    Thanks. Ok, so what you're saying is that even with the "response to resource failure" set to do not restart, the nodes will all be removed from the cluster and the vms wil be powered off.

    So, i take it there is no workaround for 2012 R2 based on your answer. (todo: upgrade to 2016 ;-) )


    This posting is provided "AS IS" with no warranties or guarantees and confers no rights

    Yep, I have tested this scenario on 2016 only... so according to Tim - there can be different behavior on 2012R2 . I think you can still remove resources from cluster and act with theme like there is no cluster in your environment but upgrade to 2016 would be better. :)

    Radek


    Thursday, April 06, 2017 9:34 AM