Hello I would like to ask a question in order to gain some opinions-knowledge, especially from Flink Contributors. Currently I am working on a task, which the proposal is to remove keyedStateOperators and use operatorState instead( CheckpointedFunction :: https://nightlies.apache.org/flink/flink-docs-release-1.17/docs/dev/datastream/fault-tolerance/state/#operator-state ) so that to reduce IO operations as the upper statement is that keyedState ones make more IO, so in order to improve the performance of the application. --- From some stress-tests I did, I did not see any big performance gains, should I have seen? --- And the second question, is in the code, some operators are keyedState and operatorState at the same time, meaning:

public final class MyFooBarProcessFunction<T> extends KeyedCoProcessFunction<String, T, FooBar, T>
      implements CheckpointedFunction

so we have some keyed state (in open(), etc.), but also some operator state, which from sustainable development is not the cleanest one, but is this a good-ok practise? or is a smell? Thanks in advance.