Shut down all VM Guests to upgrade VSM/VEM to 4.0(4)SV1(2)?

[cts_jay]

The upgrade guide for 4.0(4)SV1(2) has an unnerving pre-req:
"Step 1 The Server Administrator shuts down the VMs on the hosts running the VEM software."

Am I reading that right?  Before i can start the upgrade of the VSM and VEMs, I have to turn off every guest VM?!!!

the guide can be found here:
http://www.cisco.com/en/US/docs/switches/datacenter/nexus1000/sw/4_0_4_s_v_1_2/upgrade/software/guide/n1000v_upgrade_software.html

[Robert Burns]

Yes (in a sense).  There are two ways to upgrade the VEM portion.  If you use VUM then all VEM hosts will be upgraded at the same time.  For this they will need to be put into maintenance mode - for that to happen you need all the VMs powered off.

The other way to do it is manually installed the VEM software.  This will allow you to VMotion your running VMsto another host, upgrade it, get it back online, then move the VMs back and continue onto the next.

I'll suggest that we detail both of these methods in the upgrade guide as most people don't want downtime.

Keep in mind you're upgrading a switch.  For that to happen you will lose connectivity to every connected network interface.  With the VEM its acting like a single switch inside the host.

Follow the manual method if you can't afford downtime.

Robert

[cts_jay]

Ah, that makes more sense.  It's too bad you can't use VUM to do one VEM and thus 1 ESX Host at a time.  Feature request?

also while i understand there is usually host network downtime expected when upgrading a switch, i generally don't see many people power off all hosts connected to a switch during an upgrade.  Instead you just understand that your hosts will experience some loss of network connectivity.  and in the case where this is not acceptable usually the hosts employ some sort of nic teaming to different physical upstream switches to avoid the loss of connectivity.  I don't think there is an equivalent in virutal switching (yet).

[lwatta]

We actually upgrade the VEM modules one at a time via VUM. I just ran through the process to be sure. Right now as as stated above the manual version is the only way to prevent an outage. I agree the documentation needs some help.

We are working to make the VUM process better.

louis

[cts_jay]

How do you use VUM to do one ESX host at a time?

[lwatta]

If VUM is installed and running when you initiate the "vmware vem upgrade proceed" we will communicate with VC and start the upgrade process. The process rolls through the ESX hosts one by one ripping out the old VEM and installing the new VEM. We do not put the hosts into maintenance mode. So VMs will not get a chance to be migrated to other ESX hosts.  Hence why we say an outage is required.

To clarify my previous post. I wanted to make sure you understood that we do not upgrade all the VEMs at the same time when using VUM. We roll through the ESX hosts one by one. The process is automatic.

If I didn't answer your question or if you have other questions let us know.

louis

[cts_jay]

ok, thanks for the clarification.  Do you follow this procedure when your VSM VMs are included on ESX hosts in the cluster?

[lwatta]

I would use the manual method if my VSMs were on the ESX hosts that needed to be upgraded.

I'll run through the scenario to see if there are any gotchas, but I'm thinking you'll be fine as long as the VSM VMs are not on the host you are upgrading. So use Vmotion to move them around.