Status

Current state: Under Discussion

Discussion thread: http://apache-flink-mailing-list-archive.1008284.n3.nabble.com/DISCUSS-FLIP-70-Support-Computed-Column-for-Flink-SQL-td33126.html

...

...

Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).

...

create table T2(
  a int,
  b as a + 1 virtual,
  c bigint,
  d as c - 1 stored,
  e varchar
) with (
  k1=v1,
  k2=v2
);

...

If T2 is used as a table scan(table source), the T2 column b would be computed in a projection right above the scan, the column d comes directly from the table T2 because it is stored. That means, we would do an equivalence conversion from the scan node to a project above the scan.

...

We would put the TableColumn info of the TableSchema into the CatalogTable properties which would be used to persist. For every column expression, there are three items to persist:

• The column name
• The column data type
• The column expression of SQL-style string that looks like that how user writes them(except if the expression contains a UDF, we should expand it's schema path for a complement reference)

When deserializing, we We use SqlParser to parse the expression strings to SqlNodeinto SqlNode, then converts it to RexNode and apply the projections.

Compatibility, Deprecation, and Migration Plan

...