The MoveRangeCoordinator may return InterruptedDueToReplStateChange to the config server. When that happens, the config server marks the shard primary as unknown and blocks subsequent commands to that shard that use PrimaryOnly read preference until the next heartbeat resolves the shard's primary again (similar as what happened in SERVER-91331)

This can happen because MoveRangeCoordinator explicitly throws InterruptedDueToReplStateChange when recovering in the kMigrate and kEnterCriticalSection phases, as shown in move_range_coordinator.cpp.