Hi Dawid,
Thanks for driving this. Primary key is a very important and useful feature
to Flink Table API / SQL.
I'm +1 to the general design.
And have some thoughts as following:
1. > The design says Flink only support NOT ENFORCED mode. But the DDL
and KeyConstraint#primaryKey(..) can pass in ENFORCED mode.
What will we do when user specify an ENFORCED mode? Shall we forbid
it?
2. > "When creating a table, creating a primary key constraint will alter
the columns nullability."
I think we should force the columns to be declared NOT NULL, instead of
change the columns nullability implicitly.
Otherwise, it may be confused what is the nullbility of the primary key
columns.
3. > "Support for Unique key is not part of the FLIP. It is just mentioned
to show how can we extend the primary key concept with more constraints in
the future."
I think we can include Unique Key as part of the FLIP, as it is already
stated clearly. We can put the implemenation task of unique key in a lower
priority.
4. > Method for retrieving primary key constraint is in CatalogBaseTable.
In SQL standard, primary key is declared in line or out of line. If it
is out of line, it is still in the schema part, i.e. besides column
definitions and in the parentheses.
So I think maybe primary key should belong to `TableSchema` becuase it
is a part of schema.
CREATE TABLE xxx (
a int,
b varchar,
c bigint,
primary key (a, b)
) with (...)
Best,
Jark
On Tue, 19 Nov 2019 at 23:18, Dawid Wysakowicz <dwysakow...@apache.org>
wrote:
> Hi,
>
> I wanted to bring up the topic of primary key constraints that we could
> leverage for runtime optimizations. Please have a look at the proposal
> and I would especially draw your attention to the topic of nullability
> of columns that are part of a primary key. Some time in the future we
> will also be able to leverage it for upserting streaming tables.
>
> Here is the proposal:
>
> https://cwiki.apache.org/confluence/display/FLINK/FLIP+87%3A+Primary+key+constraints+in+Table+API
>
> Best,
>
> Dawid
>
>
>
