It's impossible to answer right off the bat without knowing more in details what's going on. HA is a complex beast, and combined with HCI, requires a lot of knowledge to find what's causing your issue, between both xha and XOSTOR.
In other words, it is very demanding to analyze all the logs and trying to make sense of it. However, I can give you some clues to make sense of it:
The HA log is at /var/log/xha.log. When you shutdown a host, you should be able to watch (on each host) what the HA is deciding to do. My gut feeling: there's maybe a XOSTOR issue making the heartbeat SR being unavailable, and so all hosts will autofence
Then you need to understand the XOSTOR logs for why the cluster wasn't doing what's expected. My best advice: remove HA first, and only then investigate on XOSTOR. Kill on node (not the master) and check if your VMs are still able to start/snapshot/write inside.
