Apache Flink

Hii all

I have a query related to `state-processor-api` which is ported to DataStream API since 1.16.0

My bootstrap job program was written in DataSet API which used to get complete in 30 mins but after migrating that to DataStream API and running it in BATCH execution mode, I am seeing a significant performance degradation. Now the same program takes *~4x* time.
I am providing same resources to both the programs.

Has anyone faced such issue ?

I am observing that `BootstrapStreamTaskRunner` acts as a blocking operator which does not start processing records until all previous tasks gets FINISHED, even though I have set `-Dexecution.batch-shuffle-mode=ALL_EXCHANGES_PIPELINED` .

This was not the case in DataSet API.

Screenshot 2023-05-03 at 12.13.33 PM.png

Screenshot 2023-05-03 at 12.38.32 PM.png

Execution with DataSet API v/s DataStream API + BATCH mode
Operator name starting with 35de is my bootstrap operator.