VMs do not migrate when etherchannel is lost

[pbsask]

We have just implemented a new node to our ESX cluster and as part of the implementation we always test failover for each redundant component.

We have a 6 nic etherchannel on each of the servers, the new server was added to the dvSwitch and each nic put into a dvUplink,  This is a 3 node cluster so each dvuplink has 3 nics one from each server.

When we where doing failure testing when one of the Nics in the etherchannel lost connectivity it displayed as link lost in vCenter, however it did not raise an alarm, which I would think it should?

We then disabled the etherchannel on the switch, which displayed all nics as having link down, my thinking here is that the host should now vmotion the VMs running on it to one of the other hosts as none of its VMs have network connectivity, however this has not happened, my VMs are still sat on the disconnected host (Management and VMotion networks are still up)

I am sure this is a config issue but I cannot for the life of me see where this might be, any help would be appreciated.

[little_horse]

DRS is not supposed to migrate VMs due to network failure  HA would restart your VMs on another node of HA cluster if its management network fails as well and your HA cluster is configured to restart VMs.  I think you need to read some guides about DRS and HA.

[pbsask]

I didnt expect DRS to do this, I did expect HA to though.  Seems silly there is no automated process to move VMs from a host that has had its VM Network fail without its management network, seems quite a plausible senario.  If you lost 75% of your nics due to a switchblade failure, personnally I would want my VMs to start migrating off so as not to bottleneck at the NICs, DRS should see this as a resource constraint, but obviously it doesn't...

In our environment we cant split the etherchannel over 2 switches (Dont know why thats what the network team said), but we split the management network and vmotion network over 2 switches, kind of makes sense to back that out and add the switch as a single point of failure so if a switch chassis fails it doesnt leave VMs on a host without a VM network......

[little_horse]

If you want your VMs to be restarted when etherchannel fails just move management VMKnic to the same vswitch where VMs reside.