In a replicaset with 4 nodes (and 1 arbiter) we had to resync completely 2 servers.
The replicaset spans accross 2 datacenters with 120ms latency.

The setup is the following :
DC 1 : nodes A and B (and arbiter)
DC 2 : nodes C and D

we stopped and clear data of B et D (A is the primary, D is perfectly up to date).

D has started to replicate but the syncingTo flag show it uses A
B has started to replicate but the syncingTo flag show it uses C

Replication is supposed to be selecting near instances, this seems broken.