are they all getting scheduled to the same node and the node is running out of memory?