It turns out to be an expected behaviour due to safety considerations. It was speculated that in the case of a total cluster split:
{1,2,3} -> (1),(2),(3)
there is a chance of operator bootstrapping node 1 into a primary component. In that case merging and recovering primary component by simple majority
(2),(3) -> {2,3}
will lead to a second PC and state divergence.
Also (and it is not a speculation) in a busy cluster a split like that may lead to one node to be a bit more updated than the others. So complete remerge of the cluster
(1),(2),(3) -> {1,2,3}
is essential here for PC recovery due to consistency considerations.
It turns out to be an expected behaviour due to safety considerations. It was speculated that in the case of a total cluster split:
{1,2,3} -> (1),(2),(3)
there is a chance of operator bootstrapping node 1 into a primary component. In that case merging and recovering primary component by simple majority
(2),(3) -> {2,3}
will lead to a second PC and state divergence.
Also (and it is not a speculation) in a busy cluster a split like that may lead to one node to be a bit more updated than the others. So complete remerge of the cluster
(1),(2),(3) -> {1,2,3}
is essential here for PC recovery due to consistency considerations.