flink-htm | streaming anomaly detection and prediction with HTM

EDIT: I just discovered that there exists a class in Java for this: NullByteKeySelector, allowing to use keyed operators for non-keyed use cases

In one of the examples of the library Flink-HTM I can read:

.keyBy { _ => None } .mapWithState { ... }

I'm trying to do the equivalent in Java. Since the mapWithState is a syntactic sugar just for Scala, I attemped to register the usual state handling a in Flink Java MappingFunction, however, I don't know how to go about the keyBy.

In this case the stream is just time series tuples of a timestamp and a numeric. I tried keying by the timestamp, the scalar, and both of them in order to enforce turning the stream into a keyedStream, so I could use keyed state as in the original example. The result was that in the next step, the registered state would be always NULL, if I understood correctly, that would make sense since the keys are constantly unique, and no partitioning is happening at all (unsure about this point).

Next thing I tried is to preserve state by switching to operator state rather than keyed state, but in this context, the state is preserved via checkpointing, checkpointing happening as defined in the checkpointing policy, which leaves me with undesired side effects at times.

I would really want to pull this together using keyed state, could somebody illustrate me on:

• what keying by "None" in the Scala version is doing, and if it possible to achieve an equivalent in Java?