Hi, I noticed that with each added (kafka) sink with exactly-once guarantees, there looks to be a penalty of ~100ms in terms of sync checkpointing time.
Would anyone be able to explain and/or point me in the right direction in the source code so that I could understand why that is? Specifically, why there appears to be a 100ms added for _each_ sink, and not a flat 100ms for all sinks, potentially pointing to a sequential set of IO calls (wiiiild guess) I would be keen to understand if there's anything I could do (incl. contributing code) that would parallelise this penalty in terms of sync checkpointing time. Alternatively, is there any setting that would help me bring the sync checkpointing time down (and still get exactly-once guarantees)? Many thanks, Fil
