I agree that it would be nice if this was stated explicitly in the docs. It’s been covered in the forum before, but it takes some digging. Maybe the new L3 docs supremo will rally the troops to move v0.5 → v1.
(Few things are more rewarding than helping those who not only wish to make things better but to also learn themselves while doing, after all, 
.)
I tried this myself, since I had a test cluster already provisioned, and it went as you said…
I was curious about what would happen when machine B woke back up, as well:
client transaction id (0000000000000005) exceeds primary transaction id (0000000000000003),
clearing client position
And then it kept chugging along (with the older database).
(I.e., no fireworks.)
Overall, you want at least two primary-candidate nodes running at all times, though, otherwise if one goes down abruptly, 
, the other won’t have the opportunity to get caught back up.