Upgrading Puppet Agent on Windows when MCO is configured and running restarts the network stack - can cause issues with Exchange servers

During an upgrade of the Puppet agent on Windows servers, when MCO is configured and running it will always cause IP Helper, DHCP and other network services to shut down and restart as part of the upgrade.

Details

There is no direct dependency on those services in Puppet agent or any of its components, they are chosen by the Microsoft MsiExec Restart Manager to restart as part of the upgrade (which means this technically falls into Microsoft's area). If we determine WHY Restart Manager is choosing to restart the network stack, we can possibly look at shutting some stuff down prior to allowing Restart Manager to make its determinations - it's a slippery slope I know. We've discussed similar issues and did some work in PA-65, which is a similar issue.

Implications

This is particularly troublesome for Exchange boxes because Exchange can be extremely finicky and if the network goes down, even for periods less than a second (possibly milliseconds), it goes into failover mode which can wreak all kinds of havoc.

Options

• Option 1 - Document this as upgrade considerations so that it becomes a well-known issue for folks upgrading. (We should probably do this anyway). That way there are no surprises that this technical limitation exists.
• Option 2 - See what workarounds we can find to overcome this limitation. We've done some work to look at possibilities with PA-65, but none of those options were any good as a manageable path forward.