HA Host Isolation Question

[nicholas1982]

Hi Guys,

I have a HA cluster with 3 ESXi 4.0 hosts. Each host has 6 NICS and 3 vDistributed Switches.

vDS 1 has 2 nics in a nic team and has a management port group configured this is the main management used for connecting to VC and can be fully resolved by DNS with FQDN

vDS 2 has 2 nics in a nic team I use this for backups and monitoring, I have also backup management port group configured just in case something goes wrong with the main management port group, I would be able to connect directely to the host via IP.

vDS 3 has 2 nics in a nic team and i use this just for vmotion traffic.

I wanted to test HA, so i disconnected one of the ESX servers uplinks for vDS 1 and sure enough that host disconnected from vCenter, but the vm hosted on that ESx host never shut down to be then powered on in on another host, however when I then disconnected the uplinks for vDS 2 the VM restarted on the other hosts in the cluster. So I guess it heartbeats through the other management port group on vDS 2 as well, but because vDS 1 is the production Network I need the host to isolate only if vDS 1 network is down, I have tried some of the advanced HA configuration without any success, has anyone got any ideas?

[a2alpha]

What you have set up and seeing is exactly as designed. You have a backup service console so that if your main one fails your vm's continue to run, which then means you don't need to invoke HA. By default the host is considered isolated if the service console (either one) can't contact their default gateway.

I would suggest either combine your first two vDS, just add all 4 uplinks to the first one. Or, you could change the isolation address it checks to being something only your first vDS can see.

However, what you are trying to do, is to con your infrastructure, because if one of your service consoles goes but the other is up, would you really want all your vm's to shut down, considering all of their networks are still up.

If you still wanted to segregate your 4 vmnics you could change the setup in the portgroup. In the first service console port group, assign vmnic0 and 1 as active the others as standby and in the second, put vmnic 2 and 3 as active and the others standby. This would still allow you to not effect backup speeds because those ports wouldn't be active for the main service console.

This is where I would start anyway, maybe have a play with it, Hope some of this helps!!

Dan

[nicholas1982]

Thanks for the reply Dan, In regards to your question, do I want the VMs to shut down if only the main network was to fail? the answer is yes because that is the network our customers use for production, its a public network every vm has a dedicated static public IP, so in the unlikely event that one of the hosts loses both vmnics to that public network, having the host shutdown the guests gracefully and power on the other hosts would be nice.

Anyway I just discovered one interesting thing which I'm still looking for documentation on, the advanced HA configuration "das.allowNetwork" would actually do the trick but for some reason in does not work on DVS only works on a standard switch.

Can anyone answer this, is the das.allowNetwork supported on Distributed Virtual Switch?

[nicholas1982]

Hi Dan,

VMware support are also working on this with me, they have asked me for the logs, they too cannot answer this. i find it really strange that no one else has had this problem, VMware fully document these HA advanced feature but no where do they say its not supported on vDS. Once I generate the logs fow VMware support I will also pass them on to you.

Nicholas

VCP4 | VTSP4 | VSP4