Memory consumption increases after ETL job

Our architecture is 3 nodes galera cluster - 2 DB nodes + 1 arbitrator in different locations. Plus running haproxy as the front end haproxy redirects traffic to one DB node in normal stage.

Users complain about the increasing memory of the 2nd node of DB when an ETL job is executed at the 1st node. The increased memory does not return to normal level. Hence, after each day of ETL job execution, more and more memory is consumed on second node and consumes all memory, resulting in a need to restart the DB on 2nd node to release the memory.

I have written 2 small stored procedures to simulate the case. Memory consumption issue is found when calling either SPs.