Amazon Managed Service for Apache Flink was previously known as Amazon Kinesis Data Analytics for Apache Flink.
I/O bound operators
It's best to avoid dependencies to external systems on the data path. It's often much more performant to keep a reference data set in state rather than querying an external system to enrich individual events. However, sometimes there are dependencies that cannot be easily moved to state, e.g., if you want to enrich events with a machine learning model that is hosted on Amazon Sagemaker.
Operators that are interfacing with external systems over the network can become a bottleneck and cause backpressure.
It is highly recommended to use AsyncIO
Moreover, for applications with I/O bound operators it can also make sense to increase the ParallelismPerKPU