Hi, I have a job with a parallelism set to 12 spread across 3 task managers (4 slots per tm). I'd like to simplify my setup and set the job parallelism to 3. I believe that we originally set the parallelism to 12 mainly for state size reasons (not really for throughput reasons except perhaps for quicker bootstrap state generation with the state-processor api). Currently the savepoints have 12 files of 260M each (~ 3G total). We use rocksdb and incremental checkpoints which seems to work well in our case with 10% (300M) transferred per checkpoints on average. Other than possible throughput degradation because of the reduced parallelism would you expect any other side-effects scaling down such a job? Thanks!

If you use Kafka you might want to read this as it discusses parallelism sizing with Kafka: https://stackoverflow.com/questions/55134009/flink-how-to-set-parallelism-properly-when-using-multiple-kafka-source