Hi team, if I want to implement an online machine learning algorithm (Stateful processing), is the table API sufficient? (I need to groupby/keyby the table source, then send each row to an algorithm). I'm confused when is the Datastream API comes into best usage in terms of stateful nature of jobs.

I’d use a simple rule: if you can implement your business logic in SQL, then the Table API would be enough for you. Otherwise, if you need a better control over state and timers use DataStream API.