Steps to reproduce: 1 Job with 2 tasks. Client + 1 driver (pool.size = 2) + 1 node. Each task (same jobUUID) is send to driver through different connection. Driver sends for execution only one task or task result is returned through different channel. This scenario is reproducible only with new load balancer. Old load balancer uses only one channel for sending all tasks. Tested workarounds:

1) generate new jobUUID for each submitted task -> job cancellation through driver MBean stops working - non existent jobUUID

2) add connection.uuid. to UuidPath (= format: <clientUUID, connection.uuid> - driver notifies 2x jobQueued, jobEnded. No tasks executed on node. This is blocker for new load balancer.

Originally it was not intended for splitting job into multiple connections. This need to be investigated after async communication with nodes will be implemented.