Adding support for postgres-xl table distribution

[Jmorjsm]

This adds support for specifying a distribution strategy (docs) when creating tables against postgres-xl.

[ 
  DISTRIBUTE BY { REPLICATION | ROUNDROBIN | { [HASH | MODULO ] ( column_name ) } } |
  DISTRIBUTED { { BY ( column_name ) } | { RANDOMLY } |
  DISTSTYLE { EVEN | KEY | ALL } DISTKEY ( column_name )
]

[roji]

@Jmorjsm thanks, that looks like a good start!

Here's a first round of review comments.

• I wouldn't worry too much about PostgreSQL-XL vs. Greenplum - the two syntaxes definitely look different enough to warrant separate APIs, even if one partially subsumes the other.
• Isn't there a missing closing curly brace in the PostgreSQL-XL spec, i.e. the one that's opened after DISTRIBUTED? I have a hard time understand which options exactly are mutually exclusive here.
• Somewhat related: the PostgreSQL-XL syntax (which I know nothing about!) seems to distinguish between DISTRIBUTE BY and DISTRIBUTED BY; the former for REPLICATION/ROUNDROBIN/<column> with/without function, the latter for RANDOMLY/<column>. Does that mean there are really two ways to define things with a column (one with DISTRIBUTE BY, one with DISTRIBUTED BY)? I don't know if any of this is significant, but ideally the builder's FluentAPI would match this the DDL syntax (so separate DistributeBy and DistributedBy)
• In theory, we should validate that there aren't two entities mapped to the same table ("table splitting") with different DISTRIBUTE BY configurations. This is pretty far-fetched, but if you want to do it, take a look at what I did in SqlServerModelValidator in https://github.com/dotnet/efcore/pull/23904/files.

[roji]

nit: can define a Deconstruct method on PostgresXlDistributeBy and use it in one line

[roji]

I was referring to the standard ValueTuple Deconstruct method as described here.

[Jmorjsm]

Thanks for your feedback here!
I did some further testing against a local postgres-xl instance and saw that the following queries were possible:

CREATE TABLE distribute_test.test_distribute_by_replication (id int, first_name varchar(100), last_name varchar(100), age int) DISTRIBUTE BY REPLICATION ;
CREATE TABLE distribute_test.test_distribute_by_roundrobin (id int, first_name varchar(100), last_name varchar(100), age int) DISTRIBUTE BY ROUNDROBIN ;
CREATE TABLE distribute_test.test_distribute_by_hash_id (id int, first_name varchar(100), last_name varchar(100), age int) DISTRIBUTE BY HASH (id);
CREATE TABLE distribute_test.test_distribute_by_modulo_id (id int, first_name varchar(100), last_name varchar(100), age int) DISTRIBUTE BY MODULO (id);

-- DOES NOT WORK: CREATE TABLE distribute_test.test_distribute_by_id (id int, first_name varchar(100), last_name varchar(100), age int) DISTRIBUTE BY (id);

CREATE TABLE distribute_test.test_distributed_by_id (id int, first_name varchar(100), last_name varchar(100), age int) DISTRIBUTED BY (id);
CREATE TABLE distribute_test.test_distributed_randomly (id int, first_name varchar(100), last_name varchar(100), age int) DISTRIBUTED RANDOMLY;

CREATE TABLE distribute_test.test_diststyle_even (id int, first_name varchar(100), last_name varchar(100), age int) DISTSTYLE EVEN;
CREATE TABLE distribute_test.test_diststyle_all (id int, first_name varchar(100), last_name varchar(100), age int) DISTSTYLE ALL;
CREATE TABLE distribute_test.test_diststyle_key_distkey_id (id int, first_name varchar(100), last_name varchar(100), age int) DISTSTYLE KEY DISTKEY (id);

Looks like DISTRIBUTED BY ({columnname}) is valid, while DISTRIBUTE BY ({columnname}) is invalid, as well as DISTSTYLE EVEN or ALL also being invalid with a distribution key.

[roji]

This can be a local method at the end of Generate(CreateTableOperation), as it's not called by anyone else.

Also, distributedByColumnName doesn't actually seem to be check anywhere.

[roji]

I've answered some questions, let me know if you're blocking on anything from my side (note there's still also my questions in #1697 (review) which may need addressing).

[roji]

Looks like DISTRIBUTED BY ({columnname}) is valid, while DISTRIBUTE BY ({columnname}) is invalid, as well as DISTSTYLE EVEN or ALL also being invalid with a distribution key.

Yeah, it seems... complicated :) But I think the fluent APIs we expose should correspond to what is actually supported on the database side.

[roji]

@Jmorjsm just to say that when this is ready for another look, please re-request a review (preferably after this passes the build too).