This is concerning me a bit:
"When a node is powered down I get a lot of errors on the cisco switch about macs flapping between ports g0/19 and g0/25."
If one of the nodes is powered down, then we can't possibly have a split-brain, so... a L2 loop??
I had the physical interfaces disappear on me once as well, after a reboot during setup and testing. KB https://kb.juniper.net/InfoCenter/index?page=content&id=KB23033&actp=search says that happens when an fxp interface has configuration. But how could that be? I didn't have any configuration, I thought, only to discover that the factory default configuration had somehow made its way back to the active config (?!?!) ...
Once the cluster was installed in production, we did experience a power loss event on one of the nodes. The packet loss was about 2-3 seconds. And it was so when the node came back online. Now I'm feeling lucky about it. Or maybe it's just that I have the two devices connected directly on control and fabric and the cluster probably reacts faster to an interface going down than it does to missed probes.