[FLINK-22612] Restructure the coders in PyFlink

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Done
• Affects Version/s: 1.14.0
• Fix Version/s: 1.14.0
• Component/s: API / Python
• Labels: pull-request-available
• Assignee: Huang Xingbo
• Reporter: Huang Xingbo
• Created: 10/May/21 02:15
• Updated: 28/Aug/21 12:15
• Resolved: 03/Jun/21 06:52

Description

Now, PyFlink has introduced many top-level coder. The top-level coder is defined as used by `Python Operation` directly. Currently, some top-level coder comes with the semantics of function, such as `TableFunctionCoder`. In fact, we need to remove this semantic and remove many coders such as `TableFunctionCoder`, `AggregateFunctionCoder` and so on. For the data structure type to be processed, there will only be `FlattenRowCoder`(used in udf, udtf.., the type of data will be List), `TopRowCoder`(used in udaf, udtagg, the type of data will be Row or InternalRow), `RawCoder`(used in datastream, the type of data will be raw data type), `ArrowCoder`(used in pandas udf, the type of data will be pandas data structure).

Issue Links

• Dependency: FLINK-22712 Support accessing Row fields by attribute name in PyFlink Row-based Operation
• links to: GitHub Pull Request #15877