1. when repl.session's statement num has reach threshold which is configured by livy.rsc.retained-statements, default as 100. The submit of 101th statement will lead to the discard of the eldest statement (should be No.1).
2. The discard means this statement is not attainable anymore. And if user forget to get the result of this discarded statement before or didn't request the result in time, he must resubmit same statement again hoping server won't discard again.
3. In above workflow, cluster do some unnecessary effort. If the discarded statement need hours of time to execute, this serve mode will waste a lot of money.
4. maybe resolve this problem by allow persist statement or discard statement that have be read in priority.