-- A bulk load writes a run of superblocks as one segment: each segment repeats the dictionaries of its columns, so the table is smaller the fewer
-- segments a load leaves. Rows of about 330 bytes are 6500 to a write buffer of 2 MB: 40000 rows are one run, one segment; 100000 rows need more.
CREATE EXTENSION pg_sift;
SET sift.table_write_buffer_limit = 2048;
CREATE TABLE twin AS SELECT g AS id, g % 7 AS grp, 'row ' || g AS v, lpad(g::text, 300, '0') AS pad FROM generate_series(1, 100000) g;
CREATE FUNCTION same_rows(a regclass, b regclass) RETURNS boolean LANGUAGE plpgsql AS $$
DECLARE r boolean;
BEGIN
	EXECUTE format('SELECT (SELECT md5(string_agg(a::text, $q$,$q$ ORDER BY id)) FROM %1$s a) = (SELECT md5(string_agg(b::text, $q$,$q$ ORDER BY id)) FROM %2$s b)', a, b) INTO r;
	RETURN r;
END $$;
CREATE TABLE small USING sift_table AS SELECT * FROM twin WHERE id <= 40000;
CREATE TABLE small_twin AS SELECT * FROM twin WHERE id <= 40000;
SELECT count(*) AS segments FROM sift.table_segments('small') WHERE kind = 'columns';
 segments 
----------
        1
(1 row)

SELECT same_rows('small', 'small_twin');
 same_rows 
-----------
 t
(1 row)

CREATE TABLE big USING sift_table AS SELECT * FROM twin;
SELECT count(*) > 1 AS several FROM sift.table_segments('big') WHERE kind = 'columns';
 several 
---------
 t
(1 row)

SELECT same_rows('big', 'twin');
 same_rows 
-----------
 t
(1 row)

DROP TABLE small, small_twin, big, twin;
DROP FUNCTION same_rows(regclass, regclass);
DROP EXTENSION pg_sift;