Could, please, anyway explain how will Flink clean-up the S3 state in case of incremental checkpointing?
For instance:
1. I put the record A into the state and keep it there for 24 hours.
2. During the next 24 hours, Flink is going to write a lot of checkpoints, but all of them will contain only the differences (delta).
3. After 24 hours, I remove the record A from state.
How is Flink going to remove record A from the S3 file that contains it? Maybe I get it wrong but it seems to me that in case of incremental checkpointing the volume of data on S3 is going to ever grow (for a single running job)