Hi All, I am facing some issues with pyflink deployment using Kubernetes Operator. I am using Map state to process the data. To store in this state, I fetch data from database for the first time when process_element is triggered for keyed stream. For this first time all the processing is completed very fast. For the second time when process_element is triggered I observed from logging that map_state.keys() is quick but when I try to map_state.get() for multiple keys, the speed is very slow and doesn't even complete most of the times. The behaviour is random as to how much time second trigger can take if at all it gets completed. I have used rocksdb backend for state. What can be the issue and how can I fix it? Thankyou for your help.
Also the system works fine when I change state to hashmap. If this can be of any help.