sharding on schema about pgcat

[levkk]

Hey @DeoLeung , give #79 a try and let me know if it works for you. Each shard now has a search_path configuration parameter, to which you can add the schema you're using in each shard; see example in CircleCI config.

That schema will be automatically findable when that shard is queried, e.g.

SET SHARD TO 1;
SELECT * FROM some_table; 
-- This will automatically look in
-- $user.some_table,
-- public.some_table,
-- and finally in shard_1.some_table

[DeoLeung]

I patch it and did a demo, which works

-- in pg 1
CREATE SCHEMA dm_1;


CREATE TABLE dm_1.test (
id INT
);

INSERT INTO dm_1.test VALUES (1), (3);

SELECT * FROM dm_1.test;

-- in pg2
CREATE SCHEMA dm_2;


CREATE TABLE dm_2.test (
id INT
);

INSERT INTO dm_2.test VALUES (2), (4);

SELECT * FROM dm_2.test;

• connect to pgcat

postgres@cb1fed2d6844:/$ psql -h pgcat -p 6432
Password for user postgres:
psql (13.7 (Debian 13.7-1.pgdg110+1))
Type "help" for help.
postgres=# show databases;
ERROR:  unrecognized configuration parameter "databases"
postgres=# SHOW DATABASES;
ERROR:  unrecognized configuration parameter "databases"
postgres=# SET SHARD TO '1';
SET SHARD
postgres=# select * from test;
 id
----
  2
  4
(2 rows)

postgres=# SET SHARD TO '0';
SET SHARD
postgres=# select * from test;
 id
----
  1
  3
(2 rows)

postgres=# select * from dm_2.test;
ERROR:  relation "dm_2.test" does not exist
LINE 1: select * from dm_2.test;
                      ^
postgres=# select * from dm_1.test;
 id
----
  1
  3
(2 rows)

postgres=# SET SHARD TO '1';
SET SHARD
postgres=# select * from dm_1.test;
ERROR:  relation "dm_1.test" does not exist
LINE 1: select * from dm_1.test;
                      ^
postgres=# select * from dm_2.test;
 id
----
  2
  4
(2 rows)

postgres=#

will start to deploy pgcat to our test cluster for some real work. further feedback may take quite a while though.
we use to dispatch our sql with full identifier(including schema), so I think we don't really need the search_path setting.
having SET SHARD to force following queries route to a specific db is enough for our scenario at the moment.

[levkk]

Great, thanks for your feedback and good luck with your test!