...
Obviously, they are the fundamental primitives as they must be provided by the framework. As for Async Processing, (Generalized) Watermark and Processing Timer Service, we need to further elaborate.
Async Processing
Flink performs user functions in a synchronous manner, but some operations need to be performed asynchronously. For example, I/O access is a time-consuming process, making the TPS for single operator much lower than in-memory computing, particularly for streaming job, when low latency is a big concern for users.
...
The dependency relationship between the API module and flink-core before and after this proposal is shown in the figure:
Umbrella FLIP is only intended to illustrate the proposed solution, so we do not want to list all the classes involved here. As for sub-FLIPs, the classes/interfaces that need to be moved to flink-core-api will be listed in detail.
...
The dependency relationship between different sub-FLIPs is shown in the following figure:
Proposed Changes
...
Compatibility, Deprecation, and Migration Plan
...