[Bug] `gp_aoseg_name` may return wrong value.

[shmiwy]

Cloudberry Database version

No response

What happened

SET gp_create_table_random_default_distribution=off;
DROP TABLE IF EXISTS foo;

CREATE TABLE foo (a INT, b INT, c CHAR(128)) WITH (appendonly=true);
CREATE INDEX foo_index ON foo(b);
INSERT INTO foo SELECT i as a, 1 as b, 'hello world' as c FROM generate_series(1, 100) AS i;

SET optimizer=off;

VACUUM foo;
DELETE FROM foo WHERE a < 4;
SELECT COUNT(*) FROM foo;

SELECT count(*) FROM pg_class WHERE relname='foo';

SELECT segno, tupcount, state FROM gp_aoseg_name('foo');
VACUUM full foo;
SELECT segno, tupcount, state FROM gp_aoseg_name('foo');
DELETE FROM foo WHERE a < 12;
SELECT segno, tupcount, state FROM gp_aoseg_name('foo'); 

drop function gp_aoseg_name;

sql with gp_aoseg_name appears to return some useless rows (tupcount = 0)

gpadmin=# SELECT COUNT(*) FROM foo;
 count
-------
    97
(1 row)

gpadmin=#
gpadmin=# SELECT count(*) FROM pg_class WHERE relname='foo';
 count
-------
     1
(1 row)

gpadmin=#
gpadmin=# SELECT segno, tupcount, state FROM gp_aoseg_name('foo');
 segno | tupcount | state
-------+----------+-------
     1 |      100 |     1
     2 |        0 |     1
     3 |        0 |     1
     4 |        0 |     1
(4 rows)

gpadmin=# VACUUM full foo;
VACUUM
gpadmin=# SELECT segno, tupcount, state FROM gp_aoseg_name('foo');
 segno | tupcount | state
-------+----------+-------
     1 |       25 |     1
     2 |       72 |     1
     3 |        0 |     1
     4 |        0 |     1
(4 rows)

gpadmin=# DELETE FROM foo WHERE a < 12;
DELETE 8
gpadmin=# SELECT segno, tupcount, state FROM gp_aoseg_name('foo');
 segno | tupcount | state
-------+----------+-------
     1 |       25 |     1
     2 |       72 |     1
     3 |        0 |     1
     4 |        0 |     1
(4 rows)

What you think should happen instead

There should be no rows with tupcount=0

How to reproduce

rerun the sql.

Operating System

centos7

Anything else

No response

Are you willing to submit PR?

• Yes, I am willing to submit a PR!

Code of Conduct

• I agree to follow this project's Code of Conduct.

[Ray-Eldath]

This is also caused by gp_appendonly_insert_files. when setting to 1 there's no lines with tupcount=0, and when setting to 8 there's seven (8-1) empty lines. should this be considered "in-fact-a-feature-not-bug" as well? @avamingli

[avamingli]

This is also caused by gp_appendonly_insert_files. when setting to 1 there's no lines with tupcount=0, and when setting to 8 there's seven (8-1) empty lines.

Agree, it's not a bug, as we will open gp_appendonly_insert_files segfiles anyway during insertion.

We could do better if there are no enough input data, the default gp_appendonly_insert_files_tuples_range is 10W,
if there are less than gp_appendonly_insert_files plus gp_appendonly_insert_files_tuples_range rows to be inserted, then we could open less segfiles than gp_appendonly_insert_files.
If it is optimized, the example of this issue could be avoided.

[avamingli]

gp_appendonly_insert_files plus gp_appendonly_insert_files_tuples_range

gp_appendonly_insert_files multiply gp_appendonly_insert_files_tuples_range

[avamingli]

This is also caused by gp_appendonly_insert_files. when setting to 1 there's no lines with tupcount=0, and when setting to 8 there's seven (8-1) empty lines.

Agree, it's not a bug, as we will open gp_appendonly_insert_files segfiles anyway during insertion.

We could do better if there are no enough input data, the default gp_appendonly_insert_files_tuples_range is 10W, if there are less than gp_appendonly_insert_files plus gp_appendonly_insert_files_tuples_range rows to be inserted, then we could open less segfiles than gp_appendonly_insert_files. If it is optimized, the example of this issue could be avoided.

I will take this one soon.

[congxuebin]

        CREATE FUNCTION gp_aoseg_name("tname" varchar)
        RETURNS TABLE("segno" integer, "tupcount" numeric, "state" smallint, "modcount" numeric) AS \$\$
            DECLARE
            aoseg_table_name varchar;
            subquery varchar;
            begin
            select concat('pg_aoseg.pg_aoseg_', array_to_string(array(select oid from pg_class where relname=tname), '')) into aoseg_table_name;
            
            subquery := 'select segno, sum(tupcount) as tupcount, state, sum(modcount) as modcount from gp_dist_random(''' || aoseg_table_name || ''') group by segno, state order by segno ';
            RETURN QUERY execute subquery;
            END\$\$
        LANGUAGE plpgsql VOLATILE
        COST 100;