If a table exists on more than one node, the selection of which node the query is routed to ends up choosing the first valid node but not the first valid node for which an open connection exists. This ends up causing problems if a node goes down but a second node would have a copy of that table. This is only of relevance when tables have redundant copies.