In serviceability you can offline the conversation manager or services on a server to gracefully have it stop taking messages, but that won’t similate a failure.
If you shut the appliance down down it should terminate any existing calls or actions. If you use SCCP “voicemail ports” they will unregister and your hunting should just drop to the other appliance. If it is a SIP trunk then it can take a bit longer depending on timers , routing , options ping etc but essentially the same thing happens.
Like with the UCM customer facing services like PCA, Web Inbox, mail relay, API (Jabber) May he pointing to one appliance or another and may be out of service. Jabber can be configured with multiple connection servers in the profiles. Mail relay will work but will come from the other host so your rules have to account for it.
Obvioisly any URLs that point to the appliance which is shutdown won’t work. When it comes back online and the cluster is restored it will replicate any messages left.
Overall there is an initial disruption but it should be alright - and this is a great reason for such a test under controlled conditions.
