Hyper-V Failover Cluster CSV Node Hiccup

[gadkins]

Hello all.

We currently use StarWind 6.0.5061 Enterprise HA Edition 4TB 2 Servers for our mirrored iSCSI SAN with both running Server 2012 (currently up to date with subscription). Performance has been great and no issues thus far have occured until here recently and was hoping you guys might shed some light or point in the right direction.

We have a three node cluster setup in Hyper-V. I went in to change the owner node that currently had the cluster shared volume to a different node as I need to bring this server down to install some more memory. Basically when I did that, the Disks listed dissappeared and did not return for roughly a minute or so (though it felt like five minutes as all of the VM's on the CSV disppeared). It did change the owner to the node I specified, but it produced some errors in the event log that stated "login request failed. the login response packet is given in the dump data."

We utilize MPIO across several nics on both the hosts and the iscsi mirrored SANs. Logging into the StarWind Software and the Events on both SANs showed the synchronization connection with the partner node was lost. I can provide more information as needed as this is pretty non technical I know.

[anton (staff)]

Sounds like a disk subsystem issues (disk is dying etc). Do you happen to have any StarWind logs?

[gadkins]

I actually went ahead and opened up support ticket since we have premium support purchased.

Initially, the engineer has ruled out it being a performance bottleneck with the disks or network configuration after reviewing everything. He mentioned that on the client side (hyper-v), moving the CSV owner node should not cause the two SANs to fall out of sync like that. I have the sync connections between the servers basically in a team of two 10 Gbps fiber connections between two different cards doing point to point. One thing he is unsure of, and is currently checking into right now, is the fact that we are using the built in teaming with Server 2012 rather than the Intel vendor based teaming mechanisms. He also mentioned that the teaming is not required for the sync connections, so we will make that change after the fact later. I believe they are going to try and reproduce my setup and see what results it produces.

[Anatoly (staff)]

Just update for the community: gadkins submitted the support case directly and currently our R&D are reviewing the logs.

[anton (staff)]

I'd like to get some results being published here. Thanks!