Hello all
I am reading a cdc-stream (strictly ordered within partition) from kafka topic (with 5 partitions).
During streaming phase, watermark advances correctly, but during backfill phase (i.e reading historical data of say last 7 days from kafka topic) watermark seems to be advance abruptly, without first processing messages with
ts
less than current watermark.
I am using partition aware watermark strategy (max-bounded-out-of-order) with
env.fromSource[Option[T]](kafkaSource, watermarkStrategy, id)
In what scenarios, this behaviour can happen ?