How does apache Beam give exactly once guarantee and do stateful calculation without checkpoint or fault tolerence?

Question

Things like groupby or combine needs exactly once guarantee for trivial calculation like sum

But apache beam seems to not have checkpoint baked in to the library, does it rely on flink or spark to manage fault tolerence and consistency in state?

Answer 1

by using Watermarks and Triggers. Watermarks are used to track the progress of event time in the data stream and Triggers are used to specify when to emit the results of a calculation. The combination of these two mechanisms allows Beam to provide exactly-once processing guarantees for stateful calculations without requiring checkpointing or fault tolerance.